UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Estimation of limb occlusion pressure by adaptation of oscillometry for surgical tourniquet control Miller, Mark E. 1989

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-UBC_1990_A7 M54.pdf [ 11.22MB ]
Metadata
JSON: 831-1.0065495.json
JSON-LD: 831-1.0065495-ld.json
RDF/XML (Pretty): 831-1.0065495-rdf.xml
RDF/JSON: 831-1.0065495-rdf.json
Turtle: 831-1.0065495-turtle.txt
N-Triples: 831-1.0065495-rdf-ntriples.txt
Original Record: 831-1.0065495-source.json
Full Text
831-1.0065495-fulltext.txt
Citation
831-1.0065495.ris

Full Text

E S T I M A T I O N O F L I M B O C C L U S I O N P R E S S U R E B Y A D A P T A T I O N O F O S C I L L O M E T R Y F O R S U R G I C A L T O U R N I Q U E T C O N T R O L Mark Edward Miller B.A.Sc. University of British Columbia, 1987 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE STUDIES DEPARTMENT OF ELECTRICAL ENGINEERING We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA December 1989 © Mark E . Miller, 1989 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. The University of British Columbia Vancouver, Canada DE-6 (2/88) ii ABSTRACT Pneumatic tourniquets are widely used in surgery of the extremities to occlude the vessels of the limb, thereby providing a bloodless field for dissection so that the surgeon may operate more quickly and accurately. Over-pressurization of the tourniquet cuff may lead to postoperative complications such as temporary or permanent paralysis of the limb. This motivated the development of adaptive tourniquet systems which could regulate the tourniquet pressure just above the limb occlusion pressure (LOP), or the minimum tourniquet pressure required to prevent blood flow past the cuff for a given duration. Previous adaptive tourniquet systems suffered from several problems which limited their practical utility in the operating room. This thesis describes the adaptation of oscillometry, a technique widely used in the noninvasive estimation of blood pressure, to the estimation of L O P in the surgical environment for application in a clinically practical adaptive tourniquet system. Improved oscillometric L O P estimation performance was obtained through the development of a filter for increasing the signal-to-noise ratio of the oscillometric pulses during periods of limb manipulation, the development of a heuristic real-time pattern recognition algorithm for extracting oscillometric pulses from signal data corrupted by limb movements, and the development of a new method for rapidly estimating the LOP which needs only one-third of the signal data required by a widely-used oscillometric approach to produce an estimate of comparable accuracy. In addition to these contributions, a new tourniquet cuff was developed which achieves an improved fit to the limb, thereby enhancing performance and reliability over that obtained from conventional tourniquets as both an oscillometric occlusion sensor and as a limb-occluding device. An adaptive tourniquet system which integrated these improvements was developed and used in a clinical study involving four orthopaedic surgeons and 16 patients. Clinical trials of the latest system version in which circumstances permitted the use of adaptive control showed that the average limb-applied pressure was reduced by 35%, or from the conventional standard of 250 mm Hg to 162 mm Hg, in the upper limb surgeries, and by 38%, or from the conventional standard of 300 mm Hg to 187 iii mm Hg, in the lower limb surgeries. These significant reductions in the pressure indicate the potential effectiveness of adaptive tourniquet control and improved cuff design on reducing the risk of patient injuries from excessive tissue compression. Furthermore, unlike all previous implementations, this system is currently being evaluated on a routine basis in orthopaedic surgical procedures performed at Vancouver General Hospital. iv TABLE OF CONTENTS Abstract List of Tables List of Figures Medical Terminology Acknowledgements xiv 1 INTRODUCTION 1 1.1 Motivation for the Research 1 1.2 Scope of the Research 2 13 Contributions of the Research 3 1.4 Thesis Overview 3 2 BACKGROUND AND REVIEW OF PREVIOUS RESEARCH 5 2.1 The Pneumatic Surgical Tourniquet 5 2.1.1 Apparatus 5 2.1 Clinical use 6 2.13 Complications associated with tourniquet use 7 2.2 Limb Occlusion Pressure Estimation Techniques 8 2.2.1 Sensor requirements for occlusion pressure estimation 8 2.2.2 Applicability of noninvasive blood pressure measurement methods to L O P estimation 9 2.2 J Oscillometry 10 2.3 Review of Previous Research in Adaptive Tourniquet Development 11 2.3.1 Characterization of the adaptive tourniquet problem 11 2.3.2 Method based on systolic pressure estimation 12 2.3 J Method based on oscillometric blood flow measurement 14 2.3.4 Method based on impedance plethysmography 15 2.4 Summary 16 3 DEVELOPMENT AND EVALUATION O F AN IMPROVED TOURNIQUET C U F F 17 3.1 Design Faults in Conventional Tourniquets 17 3.2 The Limb Occlusion Pressure Function 18 13 Anthropometric and Surgical Design Criteria 14 Implementation 15 Pressure Distributions 16 Summary 4 D E V E L O P M E N T A N D E V A L U A T I O N ©F A F I L T E R F O R SUPPRESSING M O T I O N A R T I F A C T 41 Experimental Objectives 42 Experimental Apparatus 43 Pneumatic System Transfer Function Characterization 44 Spectrum Analysis of Oscillometric and Motion Artifact Signals 4.4.1 Experimental protocol 4.4.2 Analysis of oscillometric pulses 4.43 Analysis of motion artifact 45 Suppression of Motion Artifact by Linear Processing 4.5.1 The optimum filter 4.5 Ji The near-optimum filter 4.53 Performance evaluation 46 Summary S1EVELOPMENT A N D E V A L U A T I O N S F A N A L G O R I T H M F O R E S T I M A T I N G L I M B O C C L U S I O N P R E S S U R E £1 Control Algorithm for Linear Cuff Deflation 5.1.1 Algorithm description and implementation 5.1 J: Deflation artifact inhibition 52 Detection Algorithm for Pulse Data Acquisition 5.2.1 Algorithm description 5.2.2 Algorithm implementation 5.23 Performance evaluation 53 Pulse Analysis Algorithm for LOP Estimation 53.1 Computer simulation of oscillometry 53.1.1 Model and system equations 53.1.2 Simulation experiments and results 53.13 Comparison of SBP estimation methods 53.2 Algorithm implementation 533 Performance evaluation 533.1 LOP estimation time 533.2 LOP estimation accuracy 54' Summary vi 6 DEVELOPMENT AND CLINICAL EVALUATION OF THE ADAPTIVE TOURNIQUET SYSTEM 91 6.1 System Description 91 6.1.1 Hardware 91 6.1.2 Software 94 6.2 Preparations for Clinical Trials 96 63 Experimental Protocol 97 6.4 Clinical Trial Results 98 6.4.1 Cases 1 to 3: version 1.2 software 99 6.4.2 Cases 4 to 16: version 2.1 software 101 6.4 J Synopsis 104 7 CONCLUSIONS AND RECOMMENDATIONS FOR FURTHER RESEARCH 108 7.1 Conclusions 108 7.2 Practical Improvements and Suggestions for Further Research 110 7.2.1 Practical improvements 110 7.2.1.1 Cuff design 110 7.2.1.2 Pneumatic design 110 7.2.1 J Electronic design 110 7.2.2 Topics for further research 111 7.2.2.1 Engineering research 111 7.2.2.2 Physiological research 111 REFERENCES 112 APPENDIX Patient Data File 120 LIST OF TABLES Rectangular Single-Bladder Cuff Specifications. Rectangular Dual-Bladder Cuff Specifications. Summary of Observations During Closed Loop Control of Distal Bladder Pressure in 11 Clinical Trials with Software Version 2.1. viii LIST OF FIGURES 2.1 Schematic illustrating pressure baseline and oscillometric signals during cuff deflation. 1 j 2.2 Division of adaptive tourniquet system into L O P measurement and tourniquet pressure regulation systems. 12 3.1 Raw data and least-squares limb occlusion pressure function. 19 3.2 Derivatives of limb occlusion pressure function with respect to R: (a) first derivative; (b) second derivative. 20 3.3 Tourniquet cuff application schematic. 23 3.4 Lower limb cuff assembly on patient about to undergo surgery. 24 3.5 Flexible pressure transducer. 25 3.6 Axial surface pressure distributions of cuffs on upper limb of one subject. (a) Raw distribution data of blood pressure cuff; (b) offset-corrected distributions of blood pressure cuff; (c) offset-corrected distributions of conventional dual-bladder tourniquet cuff; (d) offset-corrected distributions of new dual-bladder design. 26 3.7 Offset-corrected axial surface pressure distributions on upper limbs of three different subjects, (a) conventional dual-bladder tourniquet; (b) new dual-bladder design. 28 4.1 Pneumatic system block diagram. 32 4.2 Electronics system block diagram. 32 4.3 Pneumatic system electrical analog. 33 4.4 Pneumatic system step responses, (a) 1 meter cuff-to-computer connecting hose; (b) 3 meter connecting hose. 34 4.5 Calculated frequency responses of the pneumatic system for two hose lengths. 35 4.6 Noiseless oscillometric signal sample. 36 4.7 Motion artifact acquisition, (a) signal from distal bladder with limb at rest; (b) signal from distal bladder during flexion-extension manipulation. 37 4.8 Extracted oscillometric pulse (a) and its amplitude spectra for various values of the parameter a in the Kaiser-Bessel window: (b) a = 1.0; (c) a = 2.0; (d) a = 3.0. 39 4.9 Comparison of sensor signals: (a) oscillometric pulses; (b) moderate motion artifact; (c) high motion artifact. 40 4.10 (a) Periodogram of a windowed 10-second noise sample; (b) Welch PSD estimate of motion artifact. 42 ix 4.11 Four-pole A R models of the motion artifact PSD: (a) modeling BW = 100 Hz; (b) modeling B W = 20 Hz. 44 4.12 Power spectrum of residual e(n). 44 4.13 Oscillometric pulses corrupted by additive limb manipulation noise. 45 4.14 Signal processing model. 46 4.15 FIR filter implementation which maximizes output SNR. 47 4.16 Mean degradation from optimal SNR after analog high pass filtering. 49 4.17 (a) Motion-corrupted input signal; (b) output of inverse filter; (c) matched filter; (d) optimum filter output: convolution of (b) and (c); (e) output of analog filter. 50 4.18 Signals observed while using the oscillating pneumatic bone saw on the femur: (a) Filter input signal; (b) filter output signal. 52 4.19 Drilling the femur: (a) filter input signal; (b) filter output signal. 53 4.20 Chiseling the tibia to accept the tibial component sizer: (a) filter input signal; (b) filter output signal. 54 4.21 Inserting the femoral component of the prosthesis: (a) filter input signal; (b) filter output signal. 55 4.22 Checking the prosthesis for fit, stability, and range of motion: (a) filter input signal; (b) filter output signal. 56 5.1 Simplified electrical analog of pneumatic system. 60 5.2 Step-deflation curves obtained using a constant deflation valve-on time and valve-on time controlled by the algorithm: (a) upper limb cuff on arm; (b) lower limb cuff on leg. 61 5.3 Improvements to data acquisition electronics. 62 5.4 Step-deflation artifact observed at the amplifier output: (a) analog switch and attenuation disabled; (b) analog switch and attenuation enabled. 63 5.5 Sensor signals observed during deflation: (a) sensor baseline pressure; (b) oscillometric pulse signal. 64 5.6 Schematic illustrating detection of third pulse. 66 5.7 Real-time signal processing algorithms for pulse detection. 67 5.8 (a) Preprocessor algorithm; (b)decomposition of signal data into rising segments with local minima mini and mini+1 and local maxima max[ and maxi+]. 68 5.9 (a) Real-time pulse detection and sensor deflation control algorithm. 70 (b) Real-time pulse detection and sensor deflation control algorithm (continued). 71 5.10 Performance of pulse detection algorithm on signals corrupted by various surgical artifacts (a) Drilling artifact; (b) hammering artifact; (c) moderate motion artifact; (d) extreme motion artifact. 73 5.11 (a) Biomechanical model of cuff-limb system; (b) electrical analog of model. 77 5.12 Oscillometric pulse amplitudes obtained while varying total peripheral resistance. 80 5.13 Oscillometric pulse amplitudes obtained while varying cardiac stroke volume. 80 5.14 Linear relationships between maximum oscillometric pulse amplitude Ap and Pd = Pp and Ap and Pp = SBP - MAP. 81 5.15 (a) SBP estimation errors obtained with oscillometric data from the first simulation; (b) estimation errors obtained with oscillometric data from the second simulatioa 83 5.16 LOP estimation algorithm. 84 5.17 LOP's obtained with ultrasonic Doppler flowmeter and with estimation algorithm. 87 5.18 L O P variations observed during simulated arthroscopy of the knee. 88 5.19 Upper limb LOP variations observed in response to two Valsalva maneuvers. 89 6.1 Pneumatics of the adaptive tourniquet system. 92 6.2 Electronics of the adaptive tourniquet system. 92 6.3 The adaptive tourniquet system hardware. 94 6.4 The adaptive tourniquet system software. 95 6.5 The adaptive tourniquet system in surgery. 98 6.6 Results obtained from the first clinical trial. Surgeon: McGraw. Procedure: Total knee arthroplasty. 100 6.7 Results obtained from the sixth clinical trial. Surgeon: Beauchamp. Procedure: Total knee arthroplasty. 101 6.8 Results obtained from the seventh clinical trial. Surgeon: Beauchamp. Procedure: Internal reduction and fixation of ulna. 102 6.9 Results obtained from tenth clinical trial. Surgeon: Claridge. Procedure: Talus biopsy. 103 6.10 Results obtained from 12th clinical trial. Surgeon: Gropper. Procedure: Metacarpal arthrodesis. 104 xi MEDICAL TERMINOLOGY adduction - manipulation of a limb or joint which brings the limb closer to the midline of the body. abduction - manipulation of a limb or joint which moves the limb away from the midline of the body. arterial blood pressure - term most commonly used to describe the pressure of the blood exerted on the walls of the largest arteries. A time-varying pressure wave, the arterial blood pressure appears similar to a smoothed negative-slope sawtooth with a frequency ranging from 0.5 to 3 Hz. The exact waveshape (i.e. slope, maximum amplitude, minimum amplitude) is a function of which arterial branch the wave is observed in. arthrodesis - orthopaedic surgical procedure in which a joint is immobilized by fusion or consolidation of the joint surfaces. Literally, arthro - joint, desis - binding. arthroplasty - orthopaedic surgical procedure in which a joint is constructed, i.e. the joint is replaced by a prosthesis. Literally, arthro - joint, plasty - formation. arthroscopy - orthopaedic surgical procedure in which the interior of a joint is viewed with a needle-like optical instrument. artifact - noise or corruption in a biological signal which originates from a source not associated with the observed physiological parameter, auscultation - the act of listening to sounds made within the body for diagnostic purposes. Also refers to method most widely used by clinicians to estimate the arterial blood pressure using a blood pressure cuff and a stethoscope. In an automated instrument, the stethoscope and clinician are replaced by a vibration transducer and electronic apparatus, biopsy - the removal of tissue or other material from the living body for laboratory analysis, bpm - beats per minute; clinical term used to describe frequency of heart contraction, diastolic blood pressure (DBP) - term most commonly used to refer to the minimum value of the arterial blood pressure waveform attained during heart diastole (relaxation); cf. arterial blood pressure. distal - anatomical location further from the point of attachment of a limb to the body, dorsalis pedis artery - superficial artery palpable on top surface of foot. exsanguination - the removal of blood from within a part of the body usually by an applied force, extension - manipulation of a limb or joint which straightens the limb so that the angle between two bones is increased. x i i f e m u r - thigh bone. f lexion - manipulation of a limb or joint which folds the limb so that the angle between two bones is reduced. hypertens ive - abnormal or chronic elevation in the arterial blood pressure, hypotensive - abnormal or chronic depression in the arterial blood pressure. i n d u c t i o n - the act of initiating or inducing anesthesia or unconsciousness by administration of appropriate agents. i n t u b a t i o n - introduction of a tube through the mouth into the trachea (windpipe) which is used to carry an anesthetic-oxygen mixture to the patient, i n v a g i n a t i o n - the folding of a wall or structure in on itself to form a cavity, i s c h e m i a - deficiency of blood flow into a part of the body. l i m b occlusion pressure ( L O P ) - the minimum gas pressure within a pneumatic tourniquet required to compress the underlying vessels within the limb and thereby prevent blood flow from the body, under the tourniquet, and into the limb for a specified period of time, m e a n a r t e r i a l pressure ( M A P ) - mean value or zero-frequency component of the arterial pressure waveform; c.f. arterial blood pressure. node o f R a n v i e r - gap in organic insulator which covers the axons of some neurons to facilitate increased propagation velocity of the electrical potential along the axon, n o n i n v a s i v e - in terms of physiological measurement, any technique in which the monitored parameter is observed from outside the body, normotens ive - normal or healthy arterial blood pressure associated with particular age group and sex. o l e c r a n o n - proximal end of ulna which forms elbow; c.f. ulna. oscillometry - technique for estimating the blood pressure by observing variations in arterial volume beneath a blood pressure cuff in response to changes in the cuff pressure. These arterial volume changes are reflected as small fluctuations in the cuff pressure which vary in amplitude as the cuff is deflated. paranodal region - region on the axon of a neuron which is local to a node of Ranvier; c.f. node of Ranvier. pate l la - knee cap. proximal - anatomical location closer to the point of attachment of a limb to the body. xiii pulse pressure - the difference between the systolic blood pressure and the diastolic blood pressure; the peak-to-peak amplitude of the arterial pressure waveform, radial artery - superficial artery palpable at wrist. reduction and fixation - orthopaedic procedure in which a fracture is repaired; reduction involves alignment of the bone fragments, and fixation refers to a method of securing the bone fragments in place to facilitate healing. stroke volume (SV) - the volume of blood ejected by the left ventrical (left lower chamber) of the heart during systole (contraction), systolic blood pressure (SBP) - term most commonly used to refer to the maximum value of the arterial blood pressure waveform attained during heart systole (contraction); cf. arterial blood pressure. talus - intermediate bone between tibia (shin bone) and calcaneus (heel bone). total peripheral resistance - resistive component of the hydraulic load placed upon the heart by the entire vascular system modeled as a single element, ulna - one of the two long bones of the forearm. Valsalva maneuver - voluntary increase in thoracic pressure caused by forced exhalation against the closed glottis. x i v A C K N O W L E D G E M E N T S I would like to thank my supervisor, Dr. James McEwen, for his guidance, encouragement, and helpful suggestions throughout the course of my research, and for providing me with the unique opportunity to carry out some of my work in the operating room. I would also like to extend my gratitude to the orthopaedic surgeons, Dr. Robert McGraw, Dr. Richard Beauchamp, Dr. Richard Claridge, and Dr. Peter Gropper, without whose co-operation the clinical trials would not have been possible. Many individuals in the Dept. of Biomedical Engineering at Vancouver General Hospital assisted with the developmental aspects of the project and the experiments. Thanks to Ken Yip and Gordon McConnell for supplying the microcomputers necessary to complete this research; to Brenda Davison and Carl Janzen for helping with the assembly of the adaptive tourniquet system; to Michael Jameson for curing clunky disk drives and helping with the clinical trials; to Gordon Murphy for helping with the occlusion pressure experiment; and to the staff of the department for acting as my guinea pigs during the development of the system and the laboratory trials. Special thanks to Martine Breault of the Dept. of Orthopaedics for supplying occlusion pressure data, helping with the pressure distribution measurements, and organizing the clinical trials. Thanks are also due to Jeff Eggleston of Aspen Laboratories in Colorado, who modified the software and supplied new firmware for the ATS 1500. I would also like to thank my friends Ron Jeffery, Joel Bisson, and Kevin Pelletier for their invaluable assistance, advice, and encouragement Financial support during this research was provided by the Natural Sciences and Engineering Research Council of Canada. Chapter 1: Introduction 1 C H A P T E R 1 I N T R O D U C T I O N 1.1 Motivation for the Research Pneumatic tourniquets are used in surgery of the extremities to apply pressure to the surface of the limb, thereby collapsing the underlying vessels to prevent the flow of blood into the surgical wound. The use of a tourniquet reduces the effort of dissection, and permits improved visualization of the exposed anatomical structures so that the surgeon may operate more quickly, accurately, and atraumatically [1 - 3]. The benefits of pneumatic tourniquet use are accompanied by the risk of tissue injuries from ischemia, or deficiency of blood flow into the tissues, and from mechanical compression. Although the incidence of complications is low, the consequences of injurious compression may be exceedingly serious, resulting in temporary or permanent sensory and motor impairment from neural damage produced underneath the cuff edges [4]. Intuitively, the probability of a compression injury would be reduced if the pressure within the tourniquet could be regulated just above the limb occlusion pressure, or the minimum pressure within the tourniquet required to compress the underlying arteries and thereby prevent blood flow past the cuff for a given duration. In previous attempts to achieve this ideal condition, electronic pressure controllers were developed which adaptively adjusted the tourniquet pressure based on noninvasive estimates of time-varying physiological parameters which were related to the limb occlusion pressure. In the first of these adaptive tourniquet implementations, the physiological variable monitored was the systolic blood pressure [5], while later devices measured blood flow beneath the tourniquet cuff itself [6 - 8]. Although all of these devices could reliably track the limb occlusion pressure and could provide closed loop control of the tourniquet pressure just above this value under controlled conditions, practical problems which were unique to each implementation either precluded or severely limited their clinical use. A motivation therefore existed to improve one of these implementations such that it could be safely and effectively used in the operating room. 2 Chapter 1: Introduction 1.2 Scope of the Research The experience obtained from previous investigations in adaptive tourniquet development suggested that an approach like that used in the first implementation was most likely to succeed clinically provided that several performance limitations could be overcome. Given that the technique of noninvasively estimating the blood pressure could be directly applied or modified to estimate the limb occlusion pressure of a tourniquet cuff, the remaining issues were to develop methods for reducing limb manipulation corruption in the occlusion pressure-related signals sensed at the cuff site, detecting these signals even if they had been affected by this corruption, and increasing the speed of estimation so that rapid variations in the occlusion pressure could be tracked. Problems of this nature could be solved empirically, although a better understanding of the extent and character of the problems would be obtained if they were first approached analytically. In addition to the analytical research, practical work of a developmental nature was also required. Previous experimental research identified mechanical design faults in the tourniquet cuffs themselves [23], and reliable occlusion pressure sensing would be achieved only if the pressure within the cuff bladder could be transmitted to the limb surface predictably. This necessitated the development of a new cuff which could conform to any limb geometry, thereby attaining a perfectly snug fit along the entire cuff length. Also required was the integration of the improvements obtained from the analytical and developmental research into an adaptive tourniquet system which could be evaluated during surgical cases in the operating room. The performance of improvements in cuff design, signal processing and detection, and occlusion pressure estimation could be individually evaluated in laboratory experiments involving volunteers. The ultimate goal of the research, however, was to evaluate all of these improvements working together in a practical system during surgery, and this required the execution of a series of clinical trials. 1.3 Contributions of the Research The primary contributions of this research apply to oscillometry, a technique for noninvasively estimating the blood pressure, or in this case the limb occlusion pressure, which is described in detail in the following chapter. The most important contributions of this research were: 3 Chapter 1: Introduction i . a detailed spectral characterization of the oscillometric signal and the noise which is produced by limb movements, which led to the development and analytical comparison of optimal and near-optimal filters for suppressing the noise; i i . the development and evaluation of a unique, heuristic pattern recognition algorithm for extracting oscillometric pulse amplitudes from signal data corrupted by noise; and i i i . the development and preliminary evaluation of a new method for estimating limb occlusion pressure based on a modified form of oscillometry. This new technique facilitated a significant reduction in the quantity of signal data necessary to produce an estimate, thereby decreasing the estimation time. In addition to these contributions, the research also encompassed the development of a new tourniquet cuff which could provide improved performance and reliability over conventional tourniquets as both an oscillometric occlusion sensor and as a limb-occluding device. The implementation was based on a previously developed method of cuff application for finger tourniquets [10] which was enhanced to satisfy additional criteria on larger limbs such as patient anthropometrics and sterile field requirements for different surgical procedures. Finally, the most apparent contribution of the research was the integration of the individual improvements into a practical adaptive tourniquet system. This adaptive tourniquet was used both in the surgical treatment of patients and in the automated collection of data which quantitatively substantiated the system's performance. A clinical study illustrated that, unlike previous implementations, the system did not suffer from practical limitations which would prevent further evaluation in the operating room. The study also showed that use of the latest system version significantly reduced the average limb-applied pressure below conventional standards for every case in which circumstances permitted adaptive tourniquet control. 1.4 Thesis Overview The following chapter outlines background information related to the development of adaptive tourniquet systems and presents a review of previous research in this area. Topics discussed are the construction, clinical use, and hazards associated with conventional pneumatic tourniquets. Various techniques for nohinvasively estimating blood pressure are reviewed in terms of their applicability to the adaptive tourniquet problem, and the chapter concludes with a critical review of previous adaptive tourniquet implementations. 4 Chapter 1: Introduction Chapter 3 discusses the development of a new tourniquet cuff which achieves an improved fit to the limb which is independent of the limb geometry. Chapter 4 is a detailed investigation of the signals used in oscillometric limb occlusion pressure estimation, and examines various filters for suppressing the effects of noise introduced by limb manipulations. Chapter 5 describes the development of a limb occlusion pressure estimation algorithm, and is divided into two parts. The first part describes a pattern recognition algorithm for extracting the oscillometric signals from data corrupted by surgical noise. The second part outlines a new, faster method for estimating the limb occlusion pressure which was developed from the results of a computer simulation that modeled the biomechanical interaction of the tourniquet cuff and the limb. The results presented in the above three chapters were integrated in the development of a practical adaptive tourniquet system for use in clinical trials. The hardware and software designs of the adaptive tourniquet system are presented in Chapter 6, which also includes the results obtained with the system during sixteen surgical procedures such as total knee arthroplasty, ulna internal reduction and fixation, talus biopsy, and metacarpal arthrodesis. The chapter concludes with a synopsis which summarizes the trial results and some of the technical problems which were experienced in this clinical study. Chapter 7 reviews the contributions of the research and some of the results obtained from the experiments. The chapter also provides suggestions for practical improvements to the adaptive tourniquet system as well as topics for further investigation. Chapter 2: Background and Review of Previous Research 5 C H A P T E R 2 B A C K G R O U N D A N D R E V I E W O F P R E V I O U S R E S E A R C H This chapter outlines the design and clinical use of the pneumatic surgical tourniquet and evaluates noninvasive methods for the measurement of blood pressure in terms of their applicability to the estimation of tourniquet occlusion pressure. In particular, the oscillometric method is described and suggested to be a more appropriate choice for limb occlusion pressure estimation in the surgical environment than other methods such as auscultation. This is followed by a critical survey which discusses both the innovative aspects and the practical limitations of three previous adaptive tourniquet implementations. 2.1 The Pneumatic Surgical Tourniquet 2.1.1 Apparatus The pneumatic tourniquet consists of three main components: a cuff with an internal bladder which is wrapped circumferentially around the limb typically at its origin with the body, a regulated source of pressurized gas for inflating the cuff via one or two flexible hoses, and indicators and controls allowing the user to monitor and adjust the cuff pressure [1]. The conventional pneumatic tourniquet cuff is more rigid and generally narrower than a blood pressure cuff. In order to achieve a better fit to some limbs, the tourniquet cuff may be tapered such that it assumes a conical rather than a cylindrical shape when it is wrapped around the upper or lower extremity. Finger-sized pneumatic cuffs have also been recently introduced for use in hand surgery [11]. Control of the cuff pressure has traditionally been accomplished with mechanical pressure-regulating devices. Hazardous over-pressurization of the cuff can result from use of these pneumatic mechanisms due to excessive hysteresis in the regulator valve and offset error in the aneroid pressure gauge [12]. The microprocessor-based automatic tourniquet was introduced to eliminate this hazard by improving the accuracy, stability, and reliability of the pressure regulation [13]. This device monitors the tourniquet pressure with a solid-state transducer and pressurizes the cuff with a 6 Chapter 2: Background and Review of Previous Research solenoid/diaphragm air pump. By introducing the microprocessor to the tourniquet pressure control problem, unreliable mechanical regulators were replaced, and a foundation was established for the development of more sophisticated and safer systems such as the adaptive tourniquet, 2.1.2 Clinical use The pneumatic tourniquet was originally introduced by Cushing in 1904 for craniotomy [14], although today it is more commonly used to provide a bloodless operative field in orthopaedic procedures [1 - 3, 15 - 19]. The tourniquet may also be used to facilitate regional analgesia of the limb by preventing the release of intravenously administered anesthetics into the general circulation [15]. Usually following anesthetic induction of the patient, the cuff is snugly applied over two or three wraps of a soft, wide bandage. The non-sterile cuff is isolated from the operative site by a surgical drape, a topical anesthetic is applied to the limb, and further draping is attached. Before the cuff is inflated, the blood within the limb is removed by exsanguination. This involves elevating the limb for approximately two minutes [2, 19], or tightly wrapping the limb with an elastic Esmarch or Martin's latex bandage [1, 16, 17]. After the cuff is inflated, the latex bandage is removed. With most patients having normal circulation, the tourniquet may remain inflated for up to 90 minutes during upper limb surgery and 120 minutes during lower limb surgery without producing irreversible ischemic deterioration in the distal tissues [1 - 3, 16 - 19]. For procedures which extend beyond these time limits, it is recommended that the tourniquet be deflated for 10 to 20 minutes, depending on the change in blood pH [2]. Following this, the limb may be re-exsanguinated and the tourniquet inflated. Recommendations for tourniquet inflation pressure vary widely. Although pressures of up to 300 mm Hg have been recommended in the past for upper limb surgery [5], current texts in orthopaedic surgical practice suggest that the tourniquet should be inflated to 50 to 75 mm Hg above the preoperative systolic blood pressure [16, 17]. Because the lower limb has a larger circumference and the femoral artery is deep within it, higher pressures are required to occlude the limb. Recommendations vary from 500 mm Hg to 100 to 150 mm Hg above the systolic pressure [5, 16, 17]. One surgeon feels that a tourniquet pressure equal to three times the systolic blood pressure is required to occlude the lower limb reliably [18]. At Vancouver General Hospital, tourniquet pressures of 250 mm Hg and 300 mm Hg are routinely used in surgery of the upper and lower extremities respectively. 7 Chapter 2: Background and Review of Previous Research In many situations, these tourniquet pressures would be unnecessarily high. The above recommendations thus reflect the clinical opinion that since the limb occlusion pressure is a function of variables such as intraoperative blood pressure, limb geometry, and cuff application, a dry surgical field can be guaranteed only if the tourniquet is inflated well above this pressure. 2.13 Complications associated with tourniquet use In 1980, McEwen estimated that at least 10,000 pneumatic tourniquets were being used in over one million procedures annually in North America [3]. Given that the rate of complications has been estimated at approximately 0.2% [20], at least 2000 patients suffer injuries each year as a result of tourniquets being used in their surgical treatment. As mentioned in Chapter 1, the two main sources of these patient injuries are ischemia and mechanical compression. Ischemia causes degeneration of muscle and neural tissues underneath and distal to the cuff [1,2]. Generally, the effects of ischemia in the distal tissues are reversed if the inflation time is kept below the recommendations stated previously. Compression of the tissues may lead to muscular, vascular, cutaneous, and neurological injuries [1 - 4]. Serious neurological injuries are reflected in a postoperative loss of sensation and motor power which is usually reversible, but for some individuals these injuries may lead to a permanent disability [1]. Animal experiments involving electron microscopy of neural fibers have shown that tourniquet compression produces an axial stress which displaces the nodes of Ranvier away from the cuff site and invaginates the paranodal region furthest from the cuff in the direction of the force [4]. Only the larger diameter myelinated axons are affected, and the damage is located mosdy underneath the cuff edges where the greatest shear forces are produced [21, 22]. The extent of this type of damage is a function of the inflation pressure, and may be related to the duration of inflation. Less serious are the soft tissue injuries such as bruising, blistering, and pinching which may arise jointly from over-pressurization and improper preoperative limb preparation or cuff application technique [1,3, 12]. Under-pressurization of the tourniquet also complicates the surgery, the most apparent aspect of which is that blood will enter the surgical field. This delays the procedure and may frustrate the surgeon. In some cases, the cuff may have to be deflated and the limb exsanguinated again before the procedure can continue [1]. If the arteries remain unoccluded, passive congestion or hemorrhagic infiltration of the nerve may result [1,3, 12]. 8 Chapter 2: Background and Review of Previous Research In an adaptive tourniquet system, the limb occlusion pressure of the cuff is estimated periodically. With a reliable system, it is therefore possible to eliminate the hazards of under-pressurization while reducing the probability of a compression injury by regulating the tourniquet pressure slightly above the value required for occlusion. 2.2 Limb Occlusion Pressure Estimation Techniques 2.2.1 Sensor requirements for occlusion pressure estimation The limb occlusion pressure (LOP) of a tourniquet cuff is a time-varying quantity which is related to the intraoperative systolic blood pressure of the patient [5]. The L O P is also a function of the ratio of the cuff width to the limb circumference, the degree of agreement between the cuff and limb geometries, the snugness of cuff application, and the position of the limb with respect to the heart [5 -8, 23 - 25]. These latter effects may influence the L O P to the same extent as the systolic pressure, and this necessitates the estimation of the L O P at the site of the tourniquet cuff itself. In practical terms, an adaptive tourniquet system would ideally employ some type of occlusion-sensing apparatus either proximal to or underneath a limb-occluding tourniquet cuff. It therefore seems natural that some previous adaptive tourniquet systems have employed a dual-bladder cuff, the proximal bladder of which was cyclically inflated and deflated to estimate the LOP, and the distal bladder of which was inflated to the estimate plus a safety offset to ensure reliable occlusion [5 - 7]. Approached in this way, LOP estimation is an identical problem to the noninvasive estimation of the systolic blood pressure, and it has the same sensor requirements. For example, the ideal tourniquet L O P sensor would noninvasively respond to blood flow past the proximal bladder of the cuff, and it would be insensitive to the anatomical location of the underlying artery. It would also be mechanically robust to withstand the rigors of the orthopaedic surgical procedure, sensitive enough for use with pediatric and obese adult patients, and insensitive to noise originating from movement of the limb. A review of contemporary techniques in noninvasive blood pressure measurement shows that no sensor with all of these features is currently available [26 - 30]. The last requirement for noise immunity, or an ability to intrinsically reject motion artifact, is particularly difficult to satisfy since it conflicts with the need for sensitivity in detecting the tiny amount of energy produced by the displacement of the arterial wall. All noninvasive techniques for the measurement of blood pressure 9 Chapter 2: Background and Review of Previous Research are susceptible to motion artifacts, although some methods are marginally less sensitive than others if the motion is not excessive. 2.2.2 Applicabil i ty of noninvasive blood pressure measurement methods to L O P estimation Although most noninvasive blood pressure measurement techniques could be applied in LOP estimation, some methods are better choices than others. An auscultatory approach using a piezoelectric [31, 32] or foil electret [33] contact microphone to detect the flow-related Korotkov sounds or subaudible arterial vibrations [27 - 30, 34] would be position sensitive and possibly unable to reliably indicate flow in children and obese subjects [30]. The position-sensitivity problem could be overcome by using multiple circumferentially-distributed sensors at the expense of increased cost and susceptibility to motion artifacts. Similar problems would be experienced with the Doppler ultrasonic technique as it is especially important that the piezoelectric crystals transmitting and receiving the acoustic wave be located directly over the artery in order to detect the motion of the arterial wall due to blood flow [28]. The piezoelectric components which comprise the sensor are also more mechanically fragile than other sensors described here. Some methods of measuring blood flow or pressure would be difficult or impossible to implement at the location of the tourniquet cuff. For example, photoplethysmography, which monitors finger blood volume via observation of the amount of light from an L E D absorbed and reflected by the blood [35], would be ineffective at the tourniquet site due to the depth of the arteries in which observation of flow is required. Tonometry [29, 36] and vascular unloading of the arterial wall [26, 37], advantageous because they provide continuous monitoring of the blood pressure, are also impractical here since these techniques have been demonstrated only at superficial vessels such as the radial artery for the case of tonometry, and at the finger for the case of vascular unloading. The remaining currendy available technique is oscillometry [27, 28, 38, 39]. In this method, the bladder of the cuff itself is used as the signal sensor, thereby achieving position insensitivity and also mechanical robustness since the sensing element, a pressure transducer, is located away from the cuff within the instrument. In addition to these benefits, the method has been reported effective with neonatal, pediatric, and obese subjects [30, 39]. The primary disadvantage of the technique is that blood flow beneath the cuff is not directly observed but rather deduced from the relative amplitudes of small fluctuations in the cuff pressure. However, the accuracy of the method has been clinically 10 Chapter 2: Background and Review of Previous Research validated [38 - 40], and the technique is perhaps the most widely used method for automated monitoring because of its many practical advantages. These advantages make it a suitable choice for L O P estimation in the surgical environment. The following section will discuss the principles of the oscillometric method in detail. 2.2.3 Oscillometry Although the principles of oscillometry were reported nearly 30 years before the auscultatory method was introduced [27], only recently has this technique gained wide clinical application through the introduction of microprocessor-based instrumentation [38, 39]. Unlike other cuff-based methods in which the blood flow distal to a limb-occluding bladder is observed directly, the oscillometric technique relies on the observation of changes in the arterial volume beneath the cuff in response to variations in the externally applied pressure. These volumetric changes produce small fluctuations or "oscillations" in the cuff pressure. As the cuff is deflated from suprasystolic to subdiastolic pressures, these fluctuations initially increase in amplitude, reach a maximum of approximately 3 to 6 mm Hg, and then decrease in amplitude. In a conventional microprocessor-based instrument, the cuff pressure is transduced to an electrical signal, and the oscillometric pulses are extracted from the slowly decreasing component, or the baseline pressure, by a high pass filter and amplifier. After analog-to-digital conversion and storage of the baseline and oscillometric signals, rules are applied to the signal data to relate the pulse amplitudes, the cuff baseline pressure, and the blood pressure of the subject (Fig. 2.1). As shown in Figure 2.1, the cuff pressure at which the largest oscillometric pulse is observed is assumed equal to the mean arterial pressure (MAP) [27, 41]. The systolic blood pressure (SBP) may be estimated as the cuff pressure above M A P at which a large increase in the oscillometric pulse amplitude was observed [27, 28, 42]. Alternatively, Figure 2.1 shows that the systolic pressure may be estimated as the cuff pressure above M A P at which the pulse amplitude is approximately one-half the maximum observed amplitude [27, 39, 42]. Similarly, the diastolic blood pressure (DBP) may be estimated as the cuff pressure below M A P at which the pulse amplitude falls to 80% of the maximum observed amplitude [27, 39, 42]. Measures of the systolic pressure obtained by oscillometry correlate well with measures obtained by auscultation under controlled clinical or laboratory conditions [40]. Since the auscultatory technique actually obtains the L O P of a blood pressure cuff and equates this to the systolic pressure, 11 Chapter 2: Background and Review of Previous Research oscillometric amplitude-based rules for the estimation of systolic pressure should be applicable to the measurement of tourniquet LOP. TIME Figure 2.1: Schematic illustrating pressure baseline (upper trace) and oscillometric (lower trace) signals during cuff deflation. 2.3 Review of Previous Research in Adaptive Tourniquet Development 2.3.1 Characterization of the adaptive tourniquet problem The adaptive tourniquet problem may be characterized as two smaller subproblems with different aspects. As discussed in the previous section, one of these problems involves the measurement of L O P . This problem is complicated by issues such as estimate accuracy and resolution, estimation speed, and sensitivity of these parameters to the movement of the limb and other sources of signal corruption. Accurate estimation of the LOP would be pointless if the tourniquet pressure could not be precisely regulated at a set point determined from the estimate. This introduces a different set of problems such as regulator dynamics in response to cuff pressure changes from limb movement, shift in cuff position, and slow air leaks. Another concern is the dynamics of the pressure control system in response to changes in the set point, offset error from the set point, and hysteresis errors about the set point. Since these two problems are conceptually different, it would be best if their practical solution could be divided into two separate systems which could satisfy the requirements of each. This is illustrated in Figure 2.2. Fortunately, the dual-bladder tourniquet cuff mentioned in Section 2.2.1 Chapter 2: Background and Review of Previous Research 12 naturally lends itself to this application since the proximal bladder may be used exclusively in the measurement system and the distal bladder may be used exclusively in the limb-occluding system. Another advantage to this approach is that the regulator control problem has largely been solved with the introduction of microprocessor-based automatic tourniquets [3, 12, 13]. Limb manipulation noise Offset pressure Sensor L O P -related signal LOP measurement apparatus LOP + filtered noise Pressure generating element Tourniquet cuff Pressure transducer Measurement System Pressure Regulation System Limb manipulation noise Figure 2.2: Division of the adaptive tourniquet into LOP measurement and tourniquet pressure regulation systems. The next three sections describe previous adaptive tourniquet implementations. The innovations introduced by this research are covered, and each implementation is critically reviewed in terms of the characterization described above as well as in terms of problems which prevented the extensive clinical use of these devices. 2.3.2 Method based on systolic pressure estimation McEwen, who invented of the microprocessor-based automatic tourniquet, also originated the concept of adaptive regulation of the tourniquet pressure with McGraw in 1982 [5, 43]. Their opinion was that the tourniquet pressure should be controlled as a function of the time-varying systolic blood pressure. They were able to anticipate the effects of cuff width and application on the noninvasive estimation of this parameter and so developed a system which used a cuff with two bladders of equal width, the proximal bladder of which was connected to a commercially available automated oscillometric blood pressure monitor, and the distal bladder of which was connected to an early version of the microprocessor-based automatic tourniquet. Estimates of the systolic pressure were obtained with the monitor once per minute, and were smoothed with a three-point running average, 13 Chapter 2: Background and Review of Previous Research summed with an offset for safety, and communicated to the pressure regulator via a serial interface to control the distal bladder pressure. A significant contribution which was made in this research was that noninvasive measures of the systolic pressure could be effectively obtained proximal to an occluding tourniquet despite the significantly reduced blood flow in the region. In this work, however, McEwen and McGraw did not explicidy state that the elevated systolic pressures obtained oscillometrically from the narrow proximal bladder of the tourniquet cuff may in fact have been estimates of the cuff LOP, a term which was coined subsequent to this research. This system was effective enough to be used clinically in a limited number of trials because it optimally divided the implementation into independent measurement and pressure control systems. The practical problems which precluded further clinical use were totally associated with the measurement device, which had been originally designed to accurately estimate blood pressure in a controlled environment where limb movement could be prevented and the time required to produce an estimate was not critical. In some situations, the time between estimates became several minutes due to an inappropriate response of the instrument to periods of continuous artifact. This sensitivity to artifact and the long intervals required to obtain estimates were the primary practical limitations which prevented further clinical testing of the device. It is interesting to note that an attempt was made to commercially market an adaptive tourniquet system similar to the above implementation. The Richards Pressure Sentry (Richards Medical Co., Memphis TN) employs both a limb-occluding tourniquet cuff and a conventional blood pressure cuff. The pressure within the tourniquet, which resides on a different limb than the blood pressure cuff, is determined by summing an arbitrarily selected offset pressure to systolic blood pressure estimates obtained oscillometrically from one of the upper limbs. The primary fault with this approach is that since the tourniquet cuff and the blood pressure cuff reside on different limbs, the Pressure Sentry is insensitive to variables other than the systolic pressure which affect the L O P of the tourniquet, namely the width of the tourniquet, the snugness of the tourniquet's application, the degree of shape match between the tourniquet and the limb, and the elevation of the limb with respect to the heart. This two-cuff implementation can also be clinically inconvenient since the anesthetist requires access to one of the upper limbs to apply a blood pressure cuff for his own noninvasive monitoring equipment. These faults plus the poor ergonomic design of the Pressure Sentry made the unit unpopular with surgeons and nurses, and it is not widely used clinically. 14 Chapter 2: Background and Review of Previous Research 233 Method based on oscillometric blood flow measurement Bussani developed an ingenious method which was a hybrid of oscillometry and an auscultatory-like approach [6, 7], Again, a dual-bladder cuff was employed, although this time the distal bladder was used both to occlude the limb and to detect flow-related signals oscillometrically. The theory was that if the pressure in the proximal bladder of the cuff was greater than the LOP, blood flow would be prevented from reaching the distal bladder and producing oscillometric pressure fluctuations in it. If the proximal bladder pressure was decreased below the LOP, blood would penetrate to the distal bladder and generate an oscillometric signal there. By sweeping the proximal bladder over a narrow range of pressures, Bussani was able to achieve rapid, accurate tracking of the LOP. Bussani's opinion was that oscillometric signals could not be reliably recovered during periods of limb manipulation. In his implementation, he therefore included small bladders distal to and on top of the tourniquet cuff. These small bladders were not in the blood flow path but could detect movement of the limb. If a signal was observed in these bladders, the oscillometric signal from the tourniquet was assumed corrupted and ignored until the motion artifact completely stopped. Even if large segments of the signal data were discarded due to artifact, the method was still rapid enough to estimate L O P approximately once every 10 seconds under clinical conditions, an unprecedented achievement. Despite these advantages, the method also had serious and fundamental limitations. The most significant problem was that the distal bladder was used both as a sensor and an occluding device, and hence the measurement problem was compounded with, rather than segregated from, the pressure regulation problem. This meant that accurate regulation of the tourniquet pressure was difficult since the noise introduced by the pump and other pneumatic control elements would corrupt the oscillometric signals, which would imply a need for alternate periods of regulation and signal sensing. A second limitation was that the blood was not permitted to flow underneath the distal bladder, and hence the oscillometric signal observed was exceedingly small, perhaps only a fraction of one mm Hg in amplitude. This low signal level problem was exacerbated by the fact that the distal bladder pressure had to be maintained at a high value for reliable occlusion, thereby reducing the signal amplitude even further. Although the blood flow detection method was not dependent on the underlying arterial position, it was insensitive and thus highly susceptible to artifact, necessitating the expensive inclusion of an entire pneumatic and electronic system just to detect noise. These and other problems such as the the highly heuristic approach taken in setting the flow and noise detection thresholds suggested later that the technique would not be feasible for extensive 15 Chapter 2: Background and Review qf Previous Research clinical use. However, Bussani's system could be tuned manually with several adjustments to work with a particular subject in the clinical environment. Although closed loop control of the distal cuff pressure was never evaluated in clinical trials, extensive amounts of LOP data were acquired. Because the system could estimate LOP so quickly, previously unanticipated LOP dynamics during lower limb surgery were revealed which showed rapid, large variations in response to surgical positioning. 2.3.4 Method based on impedance plethysmography In 1988, McConnell reported a method of adaptive tourniquet control which was a radical departure from the previous techniques that had grown out of noninvasive methods for blood pressure measurement [8, 9]. In his approach, the extent of blood penetration beneath the tourniquet cuff was quantified for the first time by impedance plethysmography, a technique in which changes in blood volume due to flow may be measured by observing the pulsatile component of the electrical impedance across two surface electrodes. Using a dual-bladder tourniquet to generate a specific axial pressure distribution along the limb surface, McConnell demonstrated that the pulsatile impedance signal amplitude was related to the difference between the occlusion pressure of the cuff and the distal bladder pressure by a simple linear relationship. Using this relationship, he implemented continuous closed loop control of the cuff pressure which responded to instantaneous changes in LOP as they occurred from one heartbeat to the next McConnell's system completely integrated the processes of measurement and control rather than segregated these aspects into separate systems as suggested by Figure 2.2. Consequently, stability would be questionable during periods of limb manipulation unless approaches were implemented to reduce the sensitivity of the impedance signals to motion artifact. This question was addressed although it was not solved, and unfortunately the system could not be tested in the clinical environment. Another practical problem which complicated the use of the technique on the lower limb was an intrinsic impedance signal artifact which arose from movement of the cuff due to the large volume of blood impinging on the proximal edge. In the course of his research, McConnell revealed the highly variable nature of the axial pressure distribution produced at the limb surface by a tourniquet. He found that because of poor cuff fit, the pressure within the cuff bladder did not always correlate well with the peak surface pressure. Hence, he found the relationship between the impedance signal and the cuff pressure would not be repeatable unless the surface pressure was monitored and controlled. This was an important finding since 16 Chapter 2: Background and Review of Previous Research previously it had simply been assumed by many clinicians that the pressure within the tourniquet cuff bladder was the peak pressure transmitted to the limb surface and thus to the underlying arteries. 2.4 Summary The risks associated with pneumatic tourniquet over-pressurization and under-pressurization can be minimized through adaptively adjusting the tourniquet pressure in response to variations in the limb occlusion pressure. Oscillometry is the preferred technique for estimating the limb occlusion pressure in the clinical environment due to the positional independence and mechanical robustness of the sensor, and also due to the wide applicability of the sensing method to a variety of patients which has been reported in clinical studies. Previous adaptive tourniquet implementations which did not separate the functions of tourniquet pressure regulation and occlusion pressure estimation met with difficulties because the integration of these functions compounded the practical problems associated with each. In the implementation developed by McEwen and McGraw, estimation and regulation were performed by different systems. The only limitation with this approach, although it was an exceedingly serious one, appeared when surgical manipulation of the limb disrupted the occlusion pressure monitoring apparatus. The current research therefore focused on the development of a clinically effective adaptive tourniquet system in which limb occlusion pressure estimation was performed by an independently functioning oscillometric instrument. The objectives were to improve the performance of oscillometry in the surgical environment by examining methods of suppressing artifact, extracting information from signals corrupted by artifact, and increasing the speed of estimation. In addition to this, new dual-bladder tourniquet cuff designs were developed to improve the reliability of the pressure transmission from the cuff to the limb surface, thereby achieving increased performance as both an occlusion sensor and effector. Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff 17 CHAPTER 3 DEVELOPMENT AND EVALUATION OF AN IMPROVED TOURNIQUET CUFF The pneumatic tourniquet has been used in surgical procedures for nearly a century. Despite this, practices involving tourniquet misuse and flaws associated with the mechanical design of cuffs have only recently been identified [3, 23]. The tourniquet chosen for a given procedure is needlessly narrow in many cases, necessitating a dangerously high cuff inflation pressure to achieve reliable occlusion [25]. Furthermore, the inability of conventional cuffs to optimally adapt to different limb geometries often produces pressure distributions along the limb surface which are unpredictable in shape and magnitude, despite precise regulation of the internal bladder pressure [8, 23]. What is needed in order to achieve accurate occlusion pressure tracking and effective hemostasis is reliable transmission of the tourniquet pressure to the limb surface and therefore to the underlying vessels. This can be accomplished with a wide, compliant tourniquet cuff of inextensible material which can precisely conform to any limb shape. This chapter discusses the development of a new dual-bladder cuff which, when compared to a conventional dual-bladder tourniquet, achieves an improved fit that is independent of the limb geometry. Dimensions of the cuff have been analytically optimized to achieve the most desirable compromise between minimum limb occlusion pressure and maximum sterile field. It will be shown that the new design produces a more uniform surface pressure distribution with a much lower difference between the internal pressure and the peak surface pressure than a conventional dual-bladder tourniquet. Furthermore, these characteristics show less sensitivity to inflation pressure and limb taper with the new design than with the conventional tourniquet. As the new cuff transmits its internal pressure more predictably to the limb surface than a conventional design, improved performance and reliability in sensing and maintaining occlusion can be expected. 3.1 Design Faults in Conventional Tourniquets Because conventional tourniquets employ a nylon stiffener to prevent outward bladder expansion after inflation, they inherently assume an invariant cylindrical or conical shape when wrapped around the limb. Tourniquet geometry is set by the lengths of the cuff edges, a conical or tapered design 18 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff resulting from a distal edge which is shorter than the proximal. Due to their rigidity, conventional cuffs, especially the wider tapered designs, rarely achieve perfect conformity to human limb geometries. In the operating room, tourniquets are snugly applied to the limb such that only one finger can pass between the uninflated cuff and the limb surface [23]. Again, due to their fixed, rigid shape, it is a struggle to achieve sufficient snugness at both edges of conventional tourniquets on limbs which have large tapers. McConnell [8] and Breault [23] have both reported the deleterious effects of geometric mismatch and uncontrolled snugness on the surface pressure distribution underneath tourniquets. Generally, the degree of shape mismatch at the cuff-limb interface is reflected in the unpredictability of the pressure distribution at the proximal edge, and variations in snugness can cause the peak surface pressure to be significandy less than the internal bladder pressure. If the bladder pressure is not transmitted to the arteries underneath, errors will be obtained in the measurement of limb occlusion pressure or, in extreme cases, ineffective hemostasis may result. Although sterile disposable cuffs are now available, the expensive materials and construction techniques used in conventional tourniquets limit the wide-spread application of single-use cuffs. Because cuffs may become contaminated with blood and stained with sterilizing solution, an improved design should be less expensive than conventional disposable cuffs to permit more hospitals to adopt practices involving single use. Finally, the relationship between cuff width and limb occlusion pressure has prompted previous investgators to suggest that tourniquets should be made wider and inflated to lower pressures to reduce the risk of neural injuries [23 - 25]. The occlusion pressure-cuff width relationship and other factors affecting the optimal selection of cuff dimensions are the topics of the following two sections. 3.2 The Limb Occlusion Pressure Function In Chapter 1, the limb occlusion pressure was defined as the minimum pressure within the tourniquet which would compress the limb vessels and prevent distal blood flow over a given time period. Limb occlusion is greatly affected by the systolic blood pressure and the ratio of the cuff width to the limb circumference. This has been demonstrated for the cases of non-invasive blood pressure measurement [44] as well as in the determination of tourniquet occlusion pressure [23 - 25]. 19 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff In 1987, Breault measured the occlusion pressure of 16 different tourniquet cuffs on the upper and lower extremities of 57 subjects, both in the laboratory and the operating room [23]. The laboratory technique involved slowly deflating the tourniquet from a supraocclusive pressure until a Doppler flowmeter indicated that flow had been re-established in the radial or dorsalis pedis artery. As was done in the previous study, the raw occlusion pressure data was plotted as a function of the cuff width to limb circumference ratio, and is shown in Figure 3.1. R = Cuff width / Limb circumference Figure 3.1: Raw data [23] and least-squares limb occlusion pressure function. The method of least squares [45 , 46] was used to estimate a hyperbolic relationship between the occlusion pressure and the width-to-circumference ratio. Data from subjects with systolic blood pressure outside the range 110 - 130 mm Hg was eliminated to reduce the likelihood of hypotensive or hypertensive biases in the resulting least-squares curve, which is given by Pace = ^TT + 6 6 ' 3 3 m m H S M i l where Pocc is the hmb occlusion pressure and R is the width-to-circumference ratio. Equation (3.1), also shown in Figure 3.1, suggests that as the width-to-circumference ratio increases beyond unity, Pocc fahs from normal diastolic pressure, or about 80 mm Hg, to subdiastolic values. At the time of writing, the biomechanical causes of this effect are not completely understood. When developing a tourniquet cuff, one would like to select a bladder width as large as possible so that P 0 cc is minimized. A practical and antagonistic limitation, however, is placed upon the width 20 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff in that the surgeon must have a minimum sterile field available for dissection at, for example, the elbow or knee. This is further complicated for the cases of intravenous regional anesthesia (IVRA) [15] and the proposed adaptive tourniquet system, which both require a tourniquet with two bladders side by side. Figure 3.1 shows that as R is increased beyond 0.2, little reduction in P o c c is seen, and therefore only marginal improvement in patient safety is realized. This is further supported intuitively by inspection of the first and second derivatives of (3.1) with respect to R, illustrated in Figure 3.2, both of which relatively vanish as R approaches 0.2 from the origin. Conversely, it is clear from Figures 3.1 and 3.2 that R values less than 0.1 should not be used to avoid potentially damaging tourniquet occlusion pressures. (a) (b) Figure 3.2: Derivatives of limb occlusion pressure function with respect to R: (a) first derivative; (b) second derivative. 3.3 Anthropometric and Surgical Design Criteria Anthropometry is the statistical study of the body dimensions of large human populations. Anthropometric data is commonly used in the design of clothing, furniture, automobiles, and aircraft cockpits [47]. Several anthropometric data bases, such as the MIT Humanscale, were consulted to obtain the lengths, circumferences, and tapers of adult upper arms and thighs [47 - 49]. Both sexes, all races, and many ethnic backgrounds were included in the population statistics. Knowledge of the distributions of 21 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff limb lengths and circumferences facilitates the design of a family of tourniquets which will service a specifiable fraction of the entire population. In the proposed design, the cuff width was made equal to the limb design length minus the space required for proximal surgical exposure of the joint, with the lower bound set at twenty percent of the mean limb circumference as suggested by the limb occlusion pressure function described in Section 3.2. The limb design length in this case was specified as the mean limb length of the population the cuff was intended to service minus two standard deviations. Medical texts describing surgical exposures in orthopaedic surgery were used to quantitatively specify the surgical field required proximal to the joint [50, 51]. The lower bound on surgical field was defined as the length of the longest incision used proximal to the olecranon or the patella, regardless of the procedure, plus one centimeter for sterile preparation and draping. These dimensional investigations resulted in four rectangular cuff designs which are capable of servicing 95% of the adult male and female population in most orthopaedic procedures requiring a tourniquet. Table 3.1 lists the cuff dimensions, the range of circumferences for which the cuff is suitable, the surgical field available proximal to the joint, the range of the width-to-circumference ratio R, and the expected limb occlusion pressure obtained from (3.1) near mid-range circumference. The Cuff Bladder width (cm) Cuff length (cm) Limb circumference range (cm) Surgical field range (cm) R Expected LOP (mm Hg) Arm 1 13.2 45 20 - 30 5 - 7 0.44 - 0.66 100 Arm 2 15.2 55 30 - 45 6 - 8 0.34 -0.51 110 Leg 1 18.2 90 48 - 58 9 - 14 0.31 - 0.38 120 Leg 2 20.6 100 58 - 70 10 - 14 0.29 - 0.36 120 Table 3.1: Rectangular Single-Bladder Cuff Specifications. 22 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff cuffs are approximately 50% longer than the maximum limb circumferences for which they are intended in order to provide material for grasping during cuff application. In the adaptive tourniquet system, a cuff with two bladders is required, the proximal bladder of which acts as the occlusion sensor, and the other of which acts as the limb-occluding device. The limb occlusion pressure function (3.1) suggests that the total tourniquet width should be divided equally between the two bladders, so that an occlusion pressure measurement made with one bladder can be directly used to estimate the occlusion pressure of the other without having to introduce any scaling or offset factors to compensate for width or cuff application differences [5, 6]. Halving the bladder width, however, can significantly increase the occlusion pressure if R becomes less than 0.2 as was discussed previously. Table 3.2 lists the dimensions, the range of R, and the expected occlusion pressure for each of the four designs in Table 3.1 modified for dual bladder applications. Sub-optimal values of R less than 0.2 appear in the table. However, as the expected occlusion pressures appear to be approximately 100 mm Hg less than the tourniquet pressures used routinely on the upper and lower extremities [1, 2, 16 - 18], there still exists a motivation for adaptive control of tourniquet pressure with the proposed dual bladder cuff design. Cuff Bladder widths (cm) Cuff length (cm) Limb circumference range (cm) Surgical field range (cm) R Expected LOP (mm Hg) Arm 1 2 x 6.2 45 20 - 30 5 - 7 0.21 - 0.31 130 Arm 2 2 x 7.2 55 30 - 45 6 - 8 0.16 - 0.24 160 Leg 1 2 x 8.7 90 48 - 58 9 - 14 0.15 - 0.18 170 Leg 2 2 x 9.9 100 58 - 70 10 - 14 0.14 - 0.17 180 Table 3.2: Rectangular Dual-Bladder Cuff Specifications. 23 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff 3.4 Implementation The dual bladder tourniquet cuffs listed in Table 3.2 were constructed with polyvinyl chloride (PVC) materials using radio frequency heat sealing techniques. Upper limb cuffs were made with two layers, the outer layer being a compliant but inextensible PVC-nylon mesh composite of 10 mil (0.01 in.) thickness heat sealed to a highly compliant, extensible 10 mil transparent PVC inner layer. Lower limb cuffs, which experience greater stresses at higher inflation pressures, were made from three layers. The inner two layers were made of 20 mil clear PVC and were heat sealed to form the two bladders of the cuff. The PVC-nylon composite was then heat sealed to the bladders to form a flexible but unstretchable outer layer. No stiffener was used in these designs so that the resulting cuff would be light and capable of conforming precisely to a variety of limb geometries. Use of these materials and construction techniques resulted in a very inexpensive cuff design compared to conventional tourniquets. The method of cuff application was originated by McEwen for finger tourniquets [10, 11]. As schematically illustrated in Figure 3.3, the use of a clamp through which the cuff passes and which secures the cuff only at one point allows the tourniquet to conform to either cylindrical or conical geometries, provided that the cuff material is compliant enough. Use of such a clamping device with the compliant PVC bladders also facilitates precise adjustment of the cuff snugness. Aluminum clamps were constructed for the upper and lower limb cuffs which provided these shape and snugness adjustment features. The aluminum cuff brackets were also equipped with secondary clamps for restraining the excess cuff material from expanding upon inflation. Figure 33: Tourniquet cuff application schematic. 24 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff Figure 3.4 is a photograph of the lower limb design on a patient about to undergo a total knee replacement. The figure shows the dual P V C bladders, the cuff clamp, and the Robertson valves that were heat sealed to the bladders to provide connection to the pressure sources. The air lines from the pressure sources attach to the cuff via Luer lock connectors. Despite its width, the cuff achieves a snug fit at both proximal and distal edges. F i g u r e 3.4: Lower limb cuff assembly on patient about to undergo surgery. 3.5 Pressure Distr ibut ions The pressure distribution underneath the proposed dual bladder cuff design was measured at the limb surface using a thin, flexible pressure transducer [52]. The transducer consists of two strips of mylar heat sealed together, with a layer of metallization on each strip to create pressure-activated contact switches at discrete locations along the transducer body (Fig 3.5). The interior of the transducer is pressurized from an independent air supply such as a syringe bulb or pressure regulator. When the pressure within the transducer is equal to or greater than the surface pressure immediately over a switch pad, the electrical contact is broken. Concurrent observation of the transducer internal pressure and the status of the electrical contacts therefore provides a measure of the surface pressure distribution at the switch locations. The accuracy of this measurement technique is greatly enhanced if it is automated. A data acquisition system based on an IBM Personal Computer (IBM Corporation, Armonk NY) which had been equipped with the necessary pneumatic and A-D conversion hardware had been previously 25 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff developed by Breault for this purpose and is described in detail elsewhere [23]. This system, which provided facilities for calibrating both the internal pressure transducers and the mylar surface pressure sensors, was connected to two of the flexible sensors and was used to obtain all axial surface pressure distributions shown here. Pressure source input Luer lock connector Underside of transducer ^ — X " / U I / L _ L "X" TL Lower Upper Pressure activated switch contacts Surface metalization on mylar strip Ribbon cable connector Figure 3.5: Flexible pressure transducer. In the first experiment, a male subject with a noticeable upper limb taper (proximal circumference = 33.5 cm, distal circumference = 28 cm) was selected to obtain the axial pressure distributions underneath a standard 12 cm-wide blood pressure cuff, a conventional 15 cm-wide dual-bladder tourniquet (IVRA model 60-4006-002, Aspen Laboratories, Englewood CO), and the new 15 cm-wide upper limb design. The subject relaxed in a supine position while his right upper arm was wrapped with two layers of Webril soft bandage. Two surface pressure transducers were positioned over the biceps such that ten switch contacts could obtain the pressure distribution at nine locations spaced 1.8 cm apart. The cuff was then gendy but snugly wrapped around the limb, and, following the calibration of the surface pressure sensors by the data acquisition system, the pressure distribution was obtained at inflation pressures of 250, 200, 150, and 100 mm Hg. Particular care was taken to ensure that each cuff was applied with the same qualitative degree of snugness. For the dual-bladder cuffs, the bladders were connected in series with an aneroid pressure gauge between them to ensure that both bladders were inflated to the same pressure. Figure 3.6a shows the raw surface pressure data obtained for the blood pressure cuff and horizontal lines which indicate the actual bladder inflation pressure. This data shows that the peak surface pressure was consistently greater than the cuff pressure, a result which intuitively seems Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff 26 biased due to the known accuracy of the auscultatory method of systolic blood pressure measurement which uses this type of cuff. However, experience has shown that biased measures can be obtained with the mylar transducers if they become aged or if they are used without a lubricant. A systematic error in the results was suspected, and all subsequent surface pressure data was treated as follows to remove the bias. At each inflation pressure of the blood pressure cuff, the difference between the M l X CO 0 2.5 5 7.5 10 12.5 15 Distance from Proximal Edge (cm) (a) X 13 300 250 200 150 100 50 0 ' I I I I I -/ / / 1 / ' / / \ \ 1 A -/ ' i"~ - 1 1 / -i \ \ y / V\ -- \ ^ _ . / / i , \ i , / i , i ,^  3 0 0 -3 250 X g 200 I 150 o 100 1 oo 5 0 0 2.5 5 7.5 10 12.5 15 Distance from Proximal Edge (cm) (b) 3 0 0 -3 250 J , 200 1 150 & g 100 co 50 0 - / _\ / \ 1 1 1 S \ / / A. / / / / \ \ \ \ -\\ j 1 1 i r~~ w •.. \ / \ . N / / \ / ""\ \ \ \ l \ -lir— " W I . I . I , , V 2.5 5 7.5 10 12.5 15 Distance from Proximal Edge (cm) (C) Inflation pressure = 100 mm Hg Inflation pressure = 150 mm Hg Inflation pressure = 200 mm Hg Inflation pressure = 250 mm Hg 0 2.5 5 7.5 10 12.5 15 Distance from Proximal Edge (cm) (d) Figure 3.6: Axial surface pressure distributions of cuffs on upper l inb of one subject, (a) Raw distribution data of blood pressure cuff; (b) offset-corrected distributions of blood pressure cuff; (c) offset-corrected distributions of conventic nal dual-bladder tourniquet cuff; (d) offset-corrected distributions of new dual-bladder rjff design. Horizontal lines indicate bladder inflation pressures. 27 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff internal pressure and the mean of the three highest surface pressures was calculated to obtain pressure offset corrections of-12, -18, -24, and -24 mm Hg at cuff inflation pressures of 250, 200, 150, and 100 mm Hg respectively. These corrections were then applied to reduce the suspected systematic error in all surface pressure data. Figures 3.6b - 3.6d show the offset-corrected axial distributions of the blood pressure cuff, the conventional tourniquet, and the new design. The corrected blood pressure cuff data (Fig. 3.6b) shows close agreement between the internal pressure and the peak surface pressure, and this behavior is also reflected in the distributions produced by both bladders of the new design (Fig. 3.6d). The largest error between the internal pressure and the peak surface pressure was +11 mm Hg for the new design, and the largest difference between the peak surface pressures produced by each bladder was 3 mm Hg. As illustrated in Figure 3.6c, significantly worse results were obtained with the conventional tourniquet. From the corrected distributions, the largest error between the internal pressure and the peak surface pressure was -57 mm Hg, and the largest difference between the peak surface pressures produced by each bladder was 13 mm Hg. A comparison of Figures 3.6c and 3.6d shows that the shape of the pressure distribution of the new design varies less with inflation pressure than does the distribution of the conventional tourniquet. For example, the difference between the proximal bladder and the distal bladder peak surface pressures varied from 0 to 3 mm Hg over the range of inflation pressures. This same parameter varied from +10 to -13 mm Hg with the conventional tourniquet. The discontinuity in the distribution between the two bladders of the new design also appears proportionally smaller than the discontinuity seen with the conventional cuff, and in this respect the new design is more comparable to the blood pressure cuff than the conventional tourniquet Figure 3.7 compares the corrected axial surface pressure distributions of the conventional tourniquet cuff and the new design on the upper limbs of three subjects with different limb tapers. The bladder inflation pressure was 200 mm Hg in all cases as indicated by the horizontal line. Despite the variations in limb taper and circumference, the distributions of the new design show lower errors between the internal pressure and the peak surface pressure as well as lower inter-bladder surface pressure differences than the conventional tourniqueL These results show the importance of a snug fit along the entire length of the cuff on the reliable transmission of the internal bladder pressure to the limb surface. Because the taper of the rigid conventional tourniquet does not precisely match the limb geometry of these subjects, pressure Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff 28 Subject: M M Subject: MJ Figure 3.7: Offset-corrected axial surface pressure distributions on upper limbs of three different subjects, (a) conventional dual-bladder tourniquet; (b) new dual-bladder design. Inflation pressure is 200 mm Hg in all cases. transmission is unpredictable. This would have serious consequences on the performance of an adaptive tourniquet system. The new design provides a surface pressure distribution more comparable to that of a blood pressure cuff than the distribution of a conventional tourniquet, with much lower differences between the internal pressure and the peak surface pressure as well as lower surface pressure differences between the two bladders. 3.6 Summary Mechanical faults in the design of conventional tourniquets inspired the development of a new cuff based on limb application principles previously suggested by McEwen for finger tourniquets [10]. Original contributions in the optimal selection of dimensions based on patient anthropometrics and sterile field requirements resulted in wide designs with low predicted occlusion pressures. The design is unique in that no rigid elements are used other than a clamp which allows the same cuff to be adapted to a variety of limb shapes. Furthermore, with plastic Luer lock connectors installed, the bladder component becomes inexpensive enough to be discarded at the end of every surgical procedure. Laboratory measurements of a dual-bladder implementation revealed an axial pressure distribution which was comparable to that produced by a blood pressure cuff except for a discontinuity which appeared between the bladders. The axial pressure distribution of the new design was shown to 29 Chapter 3: Development and Evaluation of an Improved Tourniquet Cuff be less sensitive to inflation pressure and limb geometry than that of a conventional dual-bladder tourniquet. These preliminary results, although obtained from a limited population, strongly suggest that improved transmission of the cuff pressure to the limb surface will be obtained with the new cuff on most if not all patients. In terms of an adaptive tourniquet system, this improved pressure transmission implies that, with the proximal bladder acting as a oscillometric signal sensor, the new design will achieve better accuracy, repeatability, and consistency in limb occlusion pressure estimation than a conventional dual-bladder tourniquet, with increased correlation between the estimated occlusion pressure and the true occlusion pressure of the distal bladder. Furthermore, this improved pressure transmission of the new design will also result in greater reliability of the hemostasis provided by the distal bladder. Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 30 CHAPTER 4 DEVELOPMENT AND EVALUATION OF A FILTER FOR SUPPRESSING MOTION ARTIFACT A central problem in the adaptive control of tourniquet pressure lies in the reliable, noninvasive estimation of limb occlusion pressure in the noisy environment of an orthopaedic surgical procedure. No matter which noninvasive method is chosen, the estimation technique must have effective strategies for extracting useful information from signals corrupted by surgical noise, which may range from passive manipulation of the limb to the chiseling and sawing of bone to fit a prosthesis. As discussed in Chapter 2, the disruption of previous adaptive tourniquet systems by surgical manipulation, or motion artifact, limited their practical utility. This chapter discusses a linear signal processing approach to reducing the effects of motion artifact in limb occlusion pressure (LOP) estimation. Oscillometry was the technique chosen for LOP estimation primarily because of the position insensitivity and mechanical robustness of the sensing method. Oscillometric signals are the low amplitude fluctuations in the cuff pressure which are the result of changes in arterial volume beneath the cuff. This signal may be readily corrupted by artifact since the changes in limb volume associated with movement are typically much greater than the changes in arterial volume associated with the time-varying blood pressure. A detailed study of the properties of oscillometric and motion artifact signals has not previously appeared in the literature. This chapter describes the experimental acquisition and characterization of these signals using spectral estimation techniques. The resulting spectra are used in the development of an optimum digital filter which maximizes the ratio of the oscillometric signal power to the motion artifact power, defined here as the signal-to-noise ratio (SNR). The spectra are also used in an exhaustive series of discrete frequency SNR calculations to determine which of several analog filters produces the best SNR on average. Comparison of the optimal and analog filters indicates that a single analog design may replace all patient-specific optimum digital filters without a significant degradation in motion artifact rejection, thereby reducing the computational load placed upon the adaptive tourniquet system microprocessor. 31 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 4.1 Experimental Objectives Although some basic frequency domain properties of oscillometric signals have previously been reported [6, 7], the detail of the analysis was insufficient to develop optimum or near-optimum filters for motion artifact suppression To this end, the following experimental objectives were defined: i. Obtain a transfer function from the step response of the pneumatic system in order to predict the effect of system variables on linearity and signal bandwidth. ii. Acquire a data base of uncorrupted oscillometric signals from the upper and lower limbs of a highly varied population. Signals should be observed over a relatively wide bandwidth in order to obtain an accurate and complete spectral representation. iii. Acquire a data base of purely artifactual signals which contain no oscillometric components in order to obtain an estimate of the power spectral density of the noise via autoregressive analysis. iv. Compare the performance of various filter designs using output SNR as the criterion. The reasoning behind the choice of this rather than other criteria is discussed in Section 4.5. v. Verify the performance of the selected filter design both in the laboratory and the operating room. 4.2 Experimental Apparatus The experimental apparatus which was used in the signal acquisition was constructed using the pneumatic and electronic components illustrated in Figures 4.1 and 4.2. The pneumatic system (Fig. 4.1) consisted of a PVC cuff and clamp assembly similar to the upper limb dual-bladder design discussed in Chapter 3, an A T S 1000 tourniquet controller (Aspen Laboratories, Englewood CO) which acted as a regulated source of 250 mm Hg, two three-way pneumatic valves (EVO-3-3V, Clippard Instrument Laboratory, Cincinnati OH) for inflating and deflating one of the cuff bladders, a piezoresistive pressure transducer (Hewlett-Packard Co., Palo Alto CA), and 4 mm I.D. tubing for connecting the system components together. The valves were activated electrically by power M O S F E T switches which were installed in the computer. Transducer amplification was provided by the balanced-input instrumentation amplifier of the electronics system (Fig. 4.2) which produced a full output of 5 volts in response to a pressure input of 382 mm Hg at the transducer. Oscillometric signals were extracted from the cuff baseline pressure at the instrumentation amplifier output by a single pole highpass filter with cutoff frequency equal to Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 1 meter or 3 meter connecting hose IBM Personal Computer H.P. pressure V 2 AIR transducer '1 Clippard valves Electronics Aspen ATS 1000 Tourniquet Controller Aspen Dual cuff control 250 mm Hg Figure 4.1: Pneumatic system block diagram. IBM Personal Computer Pressure transducer Instrumentation amplifier 1 = Cuff baseline pressure signal 2 = Oscillometric pulse signal Gain Zero , A - D Mul t ip lexer / converter 0 - 50 X 1 / 2 High pass filter Osc. signal amp Cutoff = 0.16 Hz Upper cutoff = 105 Hz Internal Data Bus 200 Hz clock Valve controls Control logic 8 Internal Address Bus 16 Figure 4.2: Electronics system block diagram. 33 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 0.16 Hz. An amplifier with gain variable from zero to fifty and upper cutoff frequency equal to 105 Hz followed the highpass filter. As subsequent spectral analysis showed, these cutoff frequencies were much greater than the bandwidth of the signals under observation. The cuff baseline pressure and oscillometric signals were first multiplexed and then input to a 8-bit analog-to-digital converter, resulting in an acquisition system quantization resolution of 1.5 mm Hg/step for baseline signals and 0.03 mm Hg/step for oscillometric signals at maximum gain. The data sampling frequency was set by the system hardware at 200 Hz, which was well above the Nyquist rate of these signals. The transducer, valves, analog electronics, and some logic for address selection were assembled on a breadboard which was installed in the fifth expansion slot of an IBM personal computer. Programs were written in the C language to acquire the raw signal data, convert the data to floating-point format, and store the data on the fixed disk drive. 4.3 Pneumatic System Transfer Function Characterization Before proceeding with oscillometric data acquisition, the dynamic response of the pneumatic system illustrated in Figure 4.1 was measured in order to anticipate the effects of bladder interaction, bladder inflation pressure, and hose length on the signals. Similar to the fluid filled catheter systems used in invasive blood pressure measurement, the pneumatic system is best modeled as a lumped element transmission line terminated at each end by the compliances of the cuff bladder and the transducer diaphragm [53]. A consideration of the relative compliances of the elements suggests that the system may be practically modeled as the R-L-C analog shown in Figure 4.3, in which Inflate a System resistance Air inertance R L 250 mm Hg V Deflate + P(l) C System compliance Transducer Figure 43: Pneumatic system electrical analog. 34 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact resistance represents the frictional losses of the hose and valves, inductance represents the inertance of the air, and capacitance represents the compliance of the hose and cuff bladder. Briefly opening the deflation valve V2 in Figure 4.1 to air will rapidly remove some of the volume within the system, which in terms of the analog, is equivalent to switching V2 in Figure 4.3 from a to b to a again. This will excite a negative underdamped step response in the pressure p(t) observed at the transducer, the minima of which may be used to calculate the natural frequency and damping ratio of an equivalent second-order pneumatic transfer function. The P V C cuff was applied to a plexiglass cylinder 10 cm in diameter which had been wrapped with three layers of Webril soft bandage to simulate a limb. One bladder was connected to a fixed pressure supply, while the other was connected to the computer through a i m hose. The step response was obtained by inflating the monitored bladder to a given test pressure, opening the deflation valve V2 for 15 msec, and observing the a.c. component of the step response which appeared at the output of the oscillometric signal amplifier in Figure 4.2. For the purposes of these tests, the data sampling rate was increased to approximately 2 kHz. A typical response is illustrated in Figure 4.4a. = 0 E B •o 3 E < -10 -i—i—1—1—r-i—i—1—1 1 1 1—r ( t 2 , P 2 ) (tl.pl) -a x E E •a 3 E < 50 100 150 Time (msec) (a) 200 -5 -10 - (ti.pi) i I • : i J_ 50 100 150 Time (msec) (b) 200 Figure 4.4: Pneumatic system step responses, (a) 1 meter cuff-to-computer connecting hose, (b) 3 meter connecting hose. From the times and amplitudes of the first two minima shown in Figure 4.4a, the natural frequency <an of the system may be obtained from [53] 2?r (4.1) 35 Chapter 4: Development and Evaluation of a Filler for Suppressing Motion Artifact with the damping ratio £ in (4.1) given by [54] c = +(In (pi /p 2 ) ) 2 (4.2) which, for the particular case shown, results in a natural frequency of 29.6 Hz and a damping ratio of In a practical tourniquet system, a 3 m hose length is more common. As was expected, tripling the hose length increased the system damping from 0.16 to 0.23. Figure 4.4b shows, however, that the natural frequency also drops by nearly an octave to 15.8 Hz, implying a four-times increase in the inertance-compliance product. Also observable in the figure over the interval from 30 to 80 msec is a slight non-linearity which has apparently been introduced by the hose. For these reasons, use of a connecting hose longer than 3 m was avoided. With the 3 m hose attached, a total of 35 step responses were obtained at pressures of 50, 100, 150, 200 and 250 mm Hg in the measured bladder while the other bladder remained at fixed pressures of 0, 100, and 200 mm Hg. The mean natural frequency obtained was 16.2 Hz with a standard deviation of 0.28 Hz, and the mean damping ratio was 0.235 with a standard deviation of 0.016. Since the variance of the experimental results is relatively low, it can be concluded that inflation pressure has little effect on system compliance and thus on signal bandwidth, and furthermore there is little mechanical interaction between the two bladders. Figure 4.5 shows the pneumatic system second-order frequency responses calculated using the 0.16. 20 es 10 -e •20 - System with 1 m hose System with 3 m hose ' i i I i i i I i i 1 1 1 1 1 1 0 20 40 60 80 Frequency (Hz) Figure 4.5: Calculated frequency response of the pneumatic system for two hose lengths. 36 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact experimental values of con and £ for both the 1 m and 3 m hose lengths. The narrowest -3 dB bandwidth of the system is 23.5 Hz. 4.4 Spectrum Analysis of Oscillometric and Motion Artifact Signals 4.4.1 Experimental protocol Seven male and four female normotensive volunteers aged 20 to 41 were selected for acquisition of upper and lower limb oscillometric signals. Although the size of the sample was small, subjects were deliberately chosen to attain a wide qualitative variation in the ratio of fat to muscle tissue in the population. Limb circumferences ranged from 22 cm for the upper limb of a 20 year-old female to 61 cm for the thigh of a 40 year-old male. Subjects were asked to relax in a recumbent position on a patient stretcher while the 15 cm-wide dual-bladder PVC cuff was applied to the upper or lower limb. To obtain the oscillometric pulsations, the proximal bladder was connected to the data acquisition system while the distal bladder was pressurized directly from the ATS 1000 tourniquet controller. The oscillometric signal amplifier was set to its maximum gain of 50. With the distal bladder inflated to a supraocclusive pressure as it would be during surgery, the proximal bladder pressure was varied to search for the cuff baseline pressure at which the oscillometric signal had maximum amplitude. Once this pressure was determined, 3.3 seconds of signal data sampled at 200 Hz were stored to disk. A typical signal sample showing three heartbeats is illustrated in Figure 4.6. 5 a x £ E « o T3 3 "E. E < " 5 0 1 2 3 Time (sec) Figure 4.6: Noiseless oscillometric signal sample. To obtain motion artifact signals, the connections to the cuff bladders were reversed so that the proximal bladder could be inflated to a supraocclusive pressure while signal data could be acquired 37 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact from the distal bladder. The distal bladder was inflated to the same pressure as used previously to acquire the oscillometric signal pulses, and the oscillometric amplifier gain was reduced from 50 to 5 times to ensure adequate dynamic range. All floating-point signal data were later scaled to correct for changes in amplifier gain. Ten seconds of motion artifact data were acquired at a 200 Hz sampling rate while the limb was passively moved to simulate the surgical manipulations previously observed in some orthopaedic procedures. Three manipulations were performed on the limbs of each subject: flexion-extension, abduction-adduction, and a random combination of these two. Figure 4.7 shows the distal bladder signal obtained from the upper arm of a subject with the extremity at rest (Fig. 4.7a) and under moderate flexion-extension manipulation (Fig. 4.7b). The proximal bladder of the cuff arrests the blood flow in this case, and thus the use of this technique allows observation and later analysis of noise which has been isolated from the oscillometric signal. 10 -a 5 -10 I 1 I 1 I 1 I 1 I 1 I ,1,1 2 3 4 5 6 7 Time (sec) (a) 8 9 10 10 -Ti 5 ox X E E a. E < -10 -i—r—r-i—t—r—r I 1 I 1 I 1,1.1 _ [ _ 2 3 4 5 6 7 8 9 10 Time (sec) (b) Figure 4.7: Motion artifact acquisition, (a) Signal from distal bladder with limb at rest, (b) Signal from distal bladder during flexion-extension manipulation. 4.4.2 Analysis of oscillometric pulses In conventional oscillometry, the cuff pressures and the individual pulse amplitudes at those pressures are used to estimate blood pressure. Detection and estimation of oscillometric signals in noise will therefore focus on separate pulses as opposed to continuous pulse trains. For this reason, individual pulses were extracted from the uncorrupted signal data for spectrum analysis. 38 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact To reduce the effects of spectrum leakage after discrete Fourier transformation [55 - 58], each isolated pulse was multiplied by a Kaiser-Bessel data window, given by [56] lo w (n) = — na I 0 [na] 0<n<N <"> where N is the number of samples in the signal and IQ[x] is the following zero-order modified Bessel function of the first kind: io oo = £ Jk=0 (!)! (4.4) The parameter a in (4.3) allows one to compromise between the sidelobe amplitude and the main-lobe bandwidth of the window in the frequency domain. As a becomes larger, sidelobe interference is reduced at the expense of decreased spectral resolution from main-lobe convolutional smoothing. The discrete Fourier transform (DFT) of a sampled, windowed signal is defined at discrete frequencies k by [55] 1 X (*) = 7 £ M n ) * (» ) e-W ( 4 J ) J ' 71 = 0 where x(n) is the input time series and w(n) is the data window, which in this case is given by (4.3). The sampling rate fs appears in DFT above such that (4.5) approximates the continuous Fourier transform of x(t) and in fact converges to it atfs = <» [55]. The sampling rate is usually omitted from the D F T expression as it cancels out in situations where both the DFT and the inverse DFT are used. The D F T was implemented in the C language as a radix-2 fast Fourier transform based on a decimation-in-time algorithm which had been modified to accept only real-valued time series [59]. The windowed pulse data were first symmetrically padded with preceding and succeeding zeros to increase the sample length to N = 2048 points. This was done to simulate an artificial frequency resolution in the FFT which was the same as that obtained with the 2048-point noise samples. The sample mean was then subtracted from the time domain data to remove the arbitrary d.c. bias. 39 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact After obtaining the FFT, the spectrum data was divided through by the coherent gain Gw of the data window, given by 1 N ~ l G™ = JjY,™^) (4.6) n=0 This scaling corrects the spectrum for the signal attenuation introduced by the window on the time series [56]. For the rectangular or Dirichlet window, G w = 1, and for other smoother windows, Gw< 1. Figure 4.8 shows an oscillometric pulse and its amplitude spectra for values of a equal to 1.0, 2.0, and 3.0 in the windowing function (4.3). Sidelobe interference is observable in Figure 4.8b whereas Figure 4.8d shows some loss of spectral detail from the smoothing introduced by convolution with the window's main lobe. An acceptable compromise is achieved in this case for a = 2.0 as illustrated in Figure 4.8c, and thus this value was used in all subsequent windows before transformation of each Frequency (Hz) Frequency (Hz) (c) (d) Figure 4.8: Extracted oscillometric pulse (a) and its amplitude spectra for various values of parameter a in the Kaiser-Bessel window: (b) a = 1.0; (c) a = 2.0; (d) a = 3.0. 40 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact oscillometric pulse. Seventeen oscillometric pulse spectra were obtained using this approach. 4.4.3 Analysis of motion artifact Motion artifact is a random process which is assumed here to be locally wide-sense stationary. The noise can have significantly greater power than the oscillometric pulses if the manipulation of the limb is of large extent. Recovery of pulses, therefore, will most likely be successful under conditions of low and moderate artifact. For more extreme movements, saturation of the A - D converter is probable and estimation of oscillometric pulse amplitude will be impossible. Motion artifact intensity was thus quantified in terms of the cumulative energy N-l Ec=Y,V2(n) (4.7) 71=0 of the noise samples y(n). Cumulative energies ranged from 130 to 1500 cm H g 2 for the artifact signals acquired. Figure 4.9 compares an uncorrupted oscillometric pulse train with two motion artifact signals which have cumulative energies of 250 and 1300 cm H g 2 respectively. The illustration Time (sec) Figure 4.9: Comparison of sensor signals: (a) oscillometric pulses; (b) moderate motion artifact; (c) high motion artifact. 41 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact shows that large manipulations generate noise which has much greater power than the oscillometric signal, and hence a heuristic upper bound of 300 cm Hg2 was set as the maximum noise energy a practical system could effectively reject. Noise signals with energies greater than this limit were discarded, leaving 18 artifact signals for analysis. A periodogram-based approach attributed to Welch was used in obtaining the power spectral density (PSD) estimate of the noise [55, 57, 58]. Welch's method involves the division of an N-long random series y(n) into L consecutive, non-overlapping segments each of length M , multiplication of each segment by a window such as the one described by (4.3), and computation of the segment periodograms Stfk) using S i ( k ) = 1 ^ ^ - , l<i<L (4.8) where Ytfk) is the DFT of segment i given by (4.5) with x(n) = yi(n) and N = M. The PSD estimate is then obtained from the average of the periodograms as 1 1 i = l The use of windowing increases spectral smoothing at the expense of reduced resolution while periodogram averaging decreases the variance of the PSD estimate by a factor of L provided that the segments are statistically independent [57, 58]. The periodogram of a single 2048-point motion artifact sample obtained from (4.8) is shown in Figure 4.10a. The Kaiser-Bessel window (4.3) was used to multiply the time domain data with the parameter a set to 2.5. Increased values of a did not significantly improve spectral smoothing. The ensemble average (4.9) of the the 18 selected artifact periodograms produced the Welch PSD estimate illustrated in Figure 4.10b, which shows greatly reduced variance compared to the single periodogram. Autoregressive (AR) spectrum analysis was then used to obtain an all-pole, closed form expression for the PSD estimate [55, 60]. This representation was required for the implementation of the optimum filter presented in Section 4.5. An all-pole model of motion artifact is appropriate since, as Figure 4.10b shows, the noise is a low pass process with an energy concentration near the fundamental frequencies of the movement A R spectrum analysis is a least-squares approach which is based on the theory of linear prediction Chapter 4: Development and Evaluation of a Filler for Suppressing Motion Artifact 42 5 10 Frequency (Hz) (a) 15 5 10 Frequency (Hz) (b) 15 Figure 4.10: (a) Periodograra of a windowed 10 second noise sample, (b) Welch PSD estimate of motion artifact. [60]. It is assumed that the current sample of a time series y(n) may be estimated by a weighted sum of p past samples. Thus fc=i (4.10) where y(n) is the signal estimate and the a^ are called the linear predictor coefficients. Defining the residual error as e(n) = y (n) - j?(n) allows the total squared error between the signal and its estimate to be expressed as ET = £ e 2 ( n ) = £ (y(n) + ^ a k - y ( n - k)\ n n V • k=\ J (4.11) (4.12) The error may be minimized by differentiating (4.12) with respect to the coefficients a^ and setting the result equal to zero. This produces a set of p linear equations for the coefficients given by p £ ak • R (i -k)= -R(i) , 1 < i < p (4.13) fc=i Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact where R(m) is the autocorrelation of y(n) at lag m. If, as in this case, the PSD Sy(k) is available, the required autocorrelations may be obtained from the inverse DFT. Due to the symmetries involved, the following cosine transform is obtained: (N-l)/2 k=0 V 1 ' (4.14) The original signal y(n) may be written in terms of the estimate and the residual as p y(n) = y(n) + e(n)= - ak • V (n -k) + Gu (n) (4.15) where u(n) is assumed to be a white noise sequence with zero mean and unit variance. This assumption is appropriate since the PSD to be estimated in this case has the appearance of filtered white noise. Taking the z-transform of (4.15), substituting z = exp (j2nk/N), rearranging the terms and taking the squared modulus of the result produces the desired all-pole, closed form representation of the power spectrum Sy(k), given by Sy(k) ^ G2 [a - Hv (k) = ; 2 1 + £ a, • e - i2** / .v \U{k)\< (4.16) where the numerator coefficient G is found by solution of p G2 = R (0) + 52 ak • R (*0 (4.17) k=\ Since the motion artifact PSD shown in Figure 4.10b has only a single resonance at approximately 0.7 Hz, it was felt that a relatively simple AR model could be used to represent it. Figure 4.1 la shows the Welch PSD estimate and a four-pole AR model of the spectrum obtained using equations (4.13) -(4.17). The entire spectrum bandwidth of 100 Hz was used in (4.14) to obtain the autocorrelations, and Gauss-Jordan elimination [61] was used in the numerical solution of (4.13). Figure 4.11a shows that the 4-pole model has difficulty in matching the original spectrum at the lower frequencies since the concentration of the spectral power in the band 0 - 2 Hz represents only 2% of the total bandwidth over which the spectrum was modeled. Fortunately, the AR technique permits spectra to be modeled selectively with different numbers of 44 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 Frequency (Hz) Frequency (Hz) (a) (b) Figure 4.11: Four-pole A R models of the motion artifact PSD: (a) modeling BW = 100 Hz; (b) modeling BW = 20 Hz. parameters in different regions of the spectrum [60]. This is done by restricting the summation index k in (4.14) to span only the frequencies in Sy(k) where the model is to apply. This produces band-specific autocorrelation estimates R(m) which can then be used in the equations (4.13) to model the spectrum over this bandwidth. Restricting the summation (4.14) such that only frequencies in Sy(k) up to 20 Hz are included results in the improved 4-pole AR model illustrated in Figure 4.1 lb. Figure 4.12 shows that the power spectrum of the residual e(n) is nearly white in this case, indicating that a 4-pole A R model driven by a zero-mean, unit-variance white noise sequence u(n) is an adequate representation of this spectrum 0 1 2 3 4 5 6 7 8 Frequency (Hz) Figure 4.12: Power spectrum of residual error e(n). Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact over the 20 Hz bandwidth. Note, however, that solution of (4.13) assumes an effective data sampling rate of 40 Hz in this case instead of the 200 Hz rate which was actually used to acquire the noise signals. 4.5 Suppression of Motion Artifact by Linear Processing As was discussed in Section 4.4.3, practical considerations such as restricted A - D converter dynamic range limits the suppressive motion artifact power to low or moderate levels. Furthermore, as it is the pulse amplitudes which are used in oscillometric measurements, linearity must be preserved. Finally, as Figure 4.13 illustrates, passive limb manipulation results in noise which is primarily additive, as the pulses shown are still discernible after the onset of the artifact. For these reasons, a linear processing or filtering approach was selected for rejecting surgical noise. 0 1 2 3 4 5 6 7 8 9 10 Time (sec) Figure 4.13: Oscillometric pulses corrupted by additive limb manipulation noise (t > 5 seconds). The remaining question is the choice of an optimization criterion, the two most often used being minimum error in a least-squares sense (MLSE) or maximum signal-to-noise ratio (SNR). The M L S E approach results in a signal estimator such as the Weiner [63] or Kalman [62] filter, whereas the SNR approach results in an optimum processor such as the matched filter [63 - 66]. In oscillometry, the only signal parameter used is the amplitude of uniquely detected pulses. Motion artifact can corrupt these amplitudes. As preservation of the signal shape is unimportant for this application, maximization of the SNR is the best criterion to choose, which thus leads to the design of the optimum filter. 46 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 4.5.1 The optimum filter Figure 4.14 illustrates the assumed signal processing model where p(n) is an oscillometric pulse + pulse P ( n ) R p y ( k ) = 0 artifact y ( n ) x ( n ) = p ( n ) + y ( n ) v ( n) = x ( n ) * h ( n ) h ( n ) Filter Figure 4.14: Signal processing model, and y(n) is the motion artifact signal which is uncorrelated with p(n), that is, Rpy (k) = 0. The objective of the filter h(n) is to maximize the ratio of the the energy of p(n) at the filter output to the energy of the output artifact, defined as the signal-to-noise ratio (SNR). The SNR at sample instant m may be obtained from the discrete complex spectrum P(k) of pulse p(n) and the PSD Sy(k) of the artifact y(n) as follows: SNRm = (p(n)*h(n))2 = U_ E[y(n)* h(n)}2 2N N-l £ P(k)H{k)e>7*km'N k=o (N-l)/2 £ Sy(k)\H(k)\2 k=0 (4.18) The sampling rate/j appears in (4.18) in order to make the numerator and denominator summations approximate the continuous Fourier integrals from which they were obtained. The optimum filter which maximizes (4.18) is well known [63 - 66] and is obtained through algebraic manipulation of the numerator and denominator followed by application of the Schwartz inequality. The optimum filter Hopt(k) is given by Hopt (k) P- (k)e-^km"'N (4.19) where m0 is the sample time when maximum SNR occurs, which is usually chosen to ensure causality of the filter. Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 47 For white Sy(k), H0pt(k) reduces to the matched filter. If the pulse signal p(n) consists of M known samples p(0) throughp(M-l), the matched filter may be implemented as a running convolution of the input x(n) with a time-reversed and shifted version of the pulse p. In this case, the output v(n) is A/-1 v(n)= ] T x (n - k) • p (Af - 1 - k) (4.20) k=0 The matched filter for discrete signals is thus an FIR filter with M coefficients. In Section 4.4.3, an AR model of the motion artifact signal y(n) was derived. Rearranging the terms in (4.15) and taking the z-transform of the result produces E(z)= + ak-z~k Y(z) = A(z)-Y(z) (4.21) k=i The polynomial A(z) is known as the inverse filter [55, 60], which, as was shown in Figure 4.12, produces a white output E(z) in response to Y(z). Therefore, the optimum filter given by (4.19) may be practically implemented as a cascade of two digital FIR filters, the first being the inverse filter A(z), which will whiten the spectrum of x(n), followed by the matched filter of a whitened oscillometric pulse pyj(n). This filter is illustrated in Figure 4.15. P ( n ) . + x ( n ) AR PSD inverse filter P w ( n ) + e ( n ) Matched filter of P w ( n ) v ( n ) y ( n ) Figure 4.15: FIR filter implementation which maximizes output SNR. 4.5.2 The near-optimum filter The oscillometric pulse morphology is a function of patient-specific variables such as the arterial blood pressure wave shape and the ratio of limb fat to muscle tissue. Consequently, a disadvantage of the optimum filter is that is must be uniquely determined for each patient This would entail a calibration period during which uncorrupted signals would be acquired and whitened, followed by extraction and time-reversal of a single pulse for the matched filter coefficients. Another problem is 48 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact that the inverse and matched filters must be implemented digitally, which increases the computational burden on the adaptive tourniquet system microprocessor. Before A-D conversion, oscillometric signals must be extracted from the pressure baseline using a high pass filter, usually a single-pole RC network [67]. This filter also acts to suppress low frequency noise, thereby reducing A - D converter quantization distortion by allowing greater amplification to be used without increasing the likelihood of signal clipping during motion artifact. Due to the implementation problems of the optimum filter, and because some form of analog processing must precede the A - D converter, an alternative to the optimum filter based on conventional analog high pass filtering was investigated. The high pass filters examined were the single-pole RC network, the Bessel filter with 2, 3 and 4 poles, and the Butterworth filter with 2, 3 and 4 poles. From the s-domain high pass transfer functions H(s) [68], the equivalent sampled data filters H(k) were obtained using the bilinear transformation [69] >-MTT\Y The 17 oscillometric pulse spectra described in Section 4.4.2, the Welch PSD estimate obtained in Section 4.4.3, and the seven transformed high pass filters were directly substituted into (4.18). The SNR was calculated as the cutoff frequency of each filter was varied from 0.2 to 7.8 Hz in 0.2 Hz steps. The SNR degradation r\ijc was defined for each pulse spectrum Pi(k) as SNRHPJ,K VI'K=S~NR ' ( 4 ' 2 3 ) where SNRhpfjc is the SNR obtained at the output of high pass filter k with a specified cutoff frequency and SNRmaxj is the SNR of the optimum filter for pulse pi(n). The optimum filter SNR for pulse spectrum Ptfk) may be obtained by substituting (4.19) into (4.18), which yields „ , (N-l)/2 |p , M | 2 b A Umax,i - N 5y^.) The mean SNR degradation taken across the pulse spectra Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact was then computed for filter k at each of its 38 possible cutoff frequencies. Figure 4.16 illustrates the mean degradation obtained as a function of high pass cutoff frequency for the RC section, the 3-pole Bessel filter, and the 4-pole Butterworth filter. The best performance of n = -2.8 dB was attained by the 3rd order Bessel filter with a cutoff frequency of 3.2 Hz. This corresponds to a mean output SNR of +8.6dB. As a comparison, the Dinamap 845 automated oscillometric blood pressure monitor manufactured by Critikon uses a single-pole motion artifact filter with a high pass cutoff frequency of 0.9 Hz [67]. Performance of the Dinamap 845 filter is 2.7 dB worse than the Bessel filter on average. 0 i i i i | i i i i | i i i i | i i i i | i i i i | i i i i 1 i i i i. -1 -ca 2- -2 ~ e O -3 a •o -4 _ O O; R Dcgra -5 -6 ' x s. ~-Z no . 7 X c CO i : 1-poleRC o A ^ : -8 - 3-pole Bessel B -9 - 4-pole Butterworth — - A — -10 1 1 1 1 1 1 1 " 0 1 2 3 4 5 6 7 8 High Pass Filter Cutoff Frequency (Hz) Figure 4.16: Mean degradation from optimal SNR after analog high pass filtering. To complete the design of the analog filter, a low pass cutoff frequency was determined from the energy bandwidths of the pulse spectra. All spectra were found to have a 99.5% energy bandwidth less than 8 Hz, and hence the Bessel filter may be followed by an analog low pass filter with this cutoff frequency without degradation of the SNR. This is not surprising considering the results obtained for the pneumatic system bandwidth in Section 4.3. The Dinamap 845 also uses a 2-pole 8 Hz low pass filter in its oscillometric signal amplifier [67]. Use of such a filter will reduce wideband noise as well as permit a significant decrease in the sampling rate from 200 to approximately 20 Hz. 4.5 J Performance evaluation The near-optimum filter was implemented using four operational amplifiers [70]. The 3-pole Bessel filter with a high pass cutoff frequency of 3.2 Hz was realized with two active filter sections, and this was followed by a 2-pole Butterworth low pass active filter with a cutoff frequency of 8.0 Hz. The fourth operational amplifier at the end of the processing chain provided a signal gain of 44. The 50 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact optimum filter was implemented as a C language program. Figure 4.17 compares the performance of the optimum and near-optimum filters for an upper limb oscillometric signal corrupted by adduction-abduction manipulation (Fig. 4.17a). The data sampling rate for the analog signals was reduced from 200 to 40 Hz so that the 4-coefficient inverse filter, which was obtained by autoregressive modeling of the noise PSD over a restricted 20 Hz bandwidth, could be directly applied to the time series in Figure 4.17a without modification of the predictor coefficients. The output of the 4-coefficient inverse filter A(z) (Fig. 4.17b) is convolved with the matched filter of this subject (Fig. 4.17c) to produce the optimum filter output shown in Figure 4.17d. The matched (a) (b) (c) (d) (e) 0 1 2 3 4 5 6 7 8 Time (sec) Figure 4.17: (a) Motion-corrupted input signal; (b) output of inverse filter; (c) matched filter; (d) optimum filter output: convolution of (b) and (c); (e) output of analog filter (y scales are arbitrary). 51 Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact filter shown is in fact the time-reversed version of the second whitened pulse from the left which appears in the second trace (Fig. 4.17b). This pulse was extracted manually to produce the 24 matched filter coefficients since the pulse has clearly been unaffected by the motion artifact. Figure 4.17e shows the output of the analog filter. The initial assumption that moderate motion artifact is additive is validated here since, although Figure 4.17a shows that the input signal is lost in the noise for t > 4 seconds, the pulses are highly evident at the output of both linear processors. The output signals are severely distorted compared to the input signal, and this is a consequence of choosing the maximum SNR criterion as opposed to the M L S E criterion. The most important point to note is that both filters produce remarkably similar output waveforms with a practically equivalent improvement in SNR. This and the previous analytical results show that the analog filter is an effective replacement for the problematic optimum digital filters. The only type of artifact which can be simulated in the laboratory is limb manipulation. Orthopaedic surgery, however, also produces artifact from drilling, sawing, chiseling, and nailing. To evaluate the performance of the filter in the operating room, the adaptive tourniquet system described in Chapter 6 was used to acquire lower limb patient signals during a total right knee arthroplasty performed by Drs. Richard Claridge and Christopher Cameron. The distal bladder of the tourniquet was inflated to a supraocclusive pressure of 300 mm Hg and the proximal bladder, which sensed the oscillometric pulsations, was maintained at a constant pressure of 100 mm Hg. Figures 4.18 - 4.22 show a photograph of the surgical activity, the signal at the filter input, and the signal at the filter output. The signal at the filter input was obtained from the sensor baseline pressure using a filter and amplifier similar to the one used in the laboratory acquisition of pulse and noise signals. Figure 4.18 indicates that use of the pneumatic bone saw on the femur does not produce any observable corruption, probably because the saw oscillates at a frequency well above the low pass cutoff of the pneumatic system. Drilling the femur (Fig. 4.19), however, introduces a small amount of high frequency noise which is more prevalent at the filter output than at the input. Chiseling bone (Fig. 4.20) to fit the tibial prosthetic component sizer introduces high amplitude impulses (0 < t < 1.5 sec) which excite the filter and briefly saturate the amplifier. The same example shows, however, that the filter is effective in removing some motion artifact which occurs for 6 < t < 8 seconds. Figures 4.21 and 4.22 show two examples of motion artifact arising from surgical manipulations. The first example was observed when the femoral component of the prosthesis was inserted, and Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 52 Figure 4.18: Signals observed while using the oscillating pneumatic bone saw on the femur: (a) Filter input signal; (b) Filter output signal. (Courtesy Dr. R. Claridge, left, and Dr. C. Cameron, right). Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 53 —1— • * —r—i—| l —i—i—|—r—,—i—i—p—i—l l l |—! I l l | l — l l l | l l l l \ /W, 1 \ A A A A A A - 1 A / 1 (Un \\f&r Af% | i r r r L y^V rn / / n/ p^sl /^v H V ' 1 1 ' ' 1 ' 1 ' ' ' ' 1 ' ' ' ' 1 ' ' ' ' 1 ' ' ' ; 1 ' ' ' ! | I ' • I ' I ' ' ' ' 1 ' ' ' ' 1 ' ' 1 ' 1 ' 1 1 1 1 1 1 1 1 1 1 ' 1 1 1 1 ' ' ' 1 0 1 2 3 4 5 6 7 8 Time (sec) Figure 4.19: Drilling the femur: (a) Filter input signal; (b) Filter output signal. Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 54 Figure 4.20: Chiseling the tibia to accept a tibial component sizer: (a) Filter input signal; (b) Filter output signal. Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 55 0 1 2 3 4 5 6 7 8 Time (sec) Figure 4.21: Inserting the femoral component of the prosthesis: (a) Filter input signal; (b) Filter output signal. Chapter 4: Development and Evaluation of a Filter for Suppressing Motion Artifact 56 Time (sec) Figure 4.22: Checking the prosthesis for fit, stability, and range of motion: (a) Filter input signal; (b) Filter output signal. 57 Chapter 4: Development and Evaluation of a Filler for Suppressing Motion Artifact Figure 4.21 shows that the filter is effective in removing this moderate noise. Artifact of this nature occurs intermittently nearly throughout the total knee replacement. At the end of the procedure, the prosthesis is checked for fit, stability, and range of motion by repeatedly manipulating the leg from full flexion to full extension. This results in total disruption of the oscillometric signal as illustrated in Figure 4.22. Fortunately, this level of artifact is only observed over relatively brief intervals. 4.6 Summary The oscillometric waveform is a biomedical signal which has previously not been examined in detail. Step response characterization revealed that the pneumatic system bandwidth and linearity were sensitive to hose length, and a maximum length of 3 meters was specified. Spectral analysis of oscillometric pulse and motion artifact signals led to the development of optimal and near-optimal filters for motion artifact suppression. Using output SNR as a criterion, a single analog design was shown to perform nearly as well on average as all patient-specific optimum filters. Observation of the analog filter output during a total knee replacement showed that oscillometric pulses at the filter output were not seriously disrupted by sawing, drilling, or moderate limb manipulation. Use of the filter in an adaptive tourniquet system will therefore greatly improve both the speed and accuracy of limb occlusion pressure estimation. ChaplerS: Development and Evaluation of an Algorithm for Estimating LOP 58 CHAPTER 5 DEVELOPMENT AND EVALUATION OF AN ALGORITHM FOR ESTIMATING LIMB OCCLUSION PRESSURE The real-time estimation of limb occlusion pressure (LOP) in the surgical environment is significantly more difficult than the automated noninvasive measurement of blood pressure in the physician's office or intensive care unit. In the latter case, the limb is occluded for perhaps only a few seconds while the patient rests quietly during the measurement. In contrast, L O P estimation takes place in an environment where normal physiological behavior has been disrupted for an extended period of time by a local occlusion. Furthermore, the surgically treated limb does not remain at rest but is continuously acted upon in varying degrees, which corrupts the L O P estimation signals. In the physician's office, patient blood pressure is measured once or twice, and in the ICU, a new measurement is rarely required more than once every few minutes. In the operating room, however, L O P estimates must be produced once every few seconds. This last requirement, perhaps the most difficult one to achieve, is necessary to safely maintain hemostasis with variations in patient blood pressure and limb position. Before the tourniquet is inflated, anesthetic drugs have already been administered to the patient, and the patient may have been intubated. These activities can cause wide deviations in the blood pressure, but after they are completed the arterial pressure usually becomes hypotensive and stabilizes [71, 72]. However, some agents administered intravenously under anesthesia, in particular epinephrine and ephendrine [73], may induce rapid blood pressure increases of up to 50 mm Hg [74]. Instantaneous variations in L O P also arise from the surgical positioning of the limb. Using an adaptive tourniquet system which was capable of producing LOP estimates once every 12 seconds, Bussani measured occlusion pressure variations as they occurred in sympathy with lower limb manipulations [6]. His observations showed deviations of up to -40 mm Hg over a 100-second interval with limb elevation and +25 mm Hg over 50-second interval with limb depression. For these reasons, LOP estimation must be performed as quickly as possible, perhaps at the expense of precision and accuracy. Bussani's results indicate that an estimation time of 10 seconds should be practically achievable and should provide adequate LOP tracking performance. 59 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP Previous adaptive tourniquet systems were reviewed in Chapter 2. As was discussed, a dual bladder approach to adaptive tourniquet control is the preferred design solution since it separates the fundamentally different functions of LOP estimation and low hysteresis pressure regulation into two independent systems. In determining the most promising L O P estimation technique to pursue in this research, Bussani's method [6], which required alternate periods of cuff pressure regulation and signal sensing from an unresponsive flow detector, was rejected in favor of the approach developed by McEwen and McGraw using conventional oscillometry [5]. This system used an automated oscillometric monitor to obtain the systolic blood pressure (SBP) once per minute, and, although it was demonstrated that oscillometry could accurately estimate SBP in an environment with reduced blood flow, later investigations led to the opinion that the implementation was too slow and too artifact sensitive to be clinically effective. This chapter outlines the development of an improved oscillometric algorithm which is better suited to L O P estimation than conventional SBP estimation algorithms used in automated blood pressure monitors. A simple algorithm for the control of the tourniquet cuff pressure during oscillometric data acquisition is described and is shown to produce a predictable, stepwise-linear deflation characteristic which is independent of cuff application, volume, and initial inflation pressure. A heuristic pattern recognition algorithm for oscillometric pulse detection is then presented. This algorithm, which uses adaptive slope and amplitude thresholds derived from previously detected pulses, is shown to be effective in extracting usable pulses and rejecting heavily corrupted data from signal samples which were acquired during a total knee arthroplasty. A computer simulation which models the biomechanical interaction of the tourniquet cuff proximal bladder and the limb is then used to observe the oscillometric signal variations in response to changes in the total peripheral resistance of the vasculature and the cardiac stroke volume. The simulation results provide the basis for the development of a fast algorithm which requires only a few pulses obtained at low cuff pressures to produce an SBP or L O P estimate. The new algorithm is shown to produce SBP estimates which are comparable to or better than the estimates obtained from application of conventional oscillometric SBP estimation rules to the simulation output data. The chapter concludes with a laboratory performance evaluation of the new algorithm, in which LOP estimates obtained from the algorithm are compared with LOP measurements obtained with an ultrasonic Doppler flowmeter. 60 ChapterS: Development and Evaluation of an Algorithm for Estimating LOP 5.1 Control Algorithm for Linear Cuff Deflation 5.1.1 Algorithm description and implementation Nearly all noninvasive methods for the estimation of blood pressure or LOP rely on the same principle, which involves the observation of a flow-related variable while a limb occluding cuff is deflated. In traditional noninvasive blood pressure measurements, the cuff is deflated through a small orifice producing a long, exponential pressure decay which is a function of the cuff volume and initial inflation pressure. With this deflation technique, variations in limb application, cuff volume, and the initial cuff inflation pressure must be compensated by precisely controlling the diameter of the deflation orifice to achieve the same deflation rate under all conditions. Linear deflation of the cuff, however, decreases the cuff pressure more rapidly than an exponential decay over the same pressure range, thereby speeding up the data acquisition [75]. Step-deflating the cuff after each detected heartbeat permits a specifiable deflation rate to be obtained independent of the cuff volume and initial pressure by simply controlling the time that a large deflation orifice is opened to atmospheric pressure. For these reasons, an algorithm was developed for linearly step-deflating the proximal bladder of the tourniquet cuff, referred to hereafter as the sensor. The pneumatic system electrical analog presented in Section 4.3 was simplified to the circuit shown in Figure 5.1 and was used to develop a linear step-deflation algorithm for the tourniquet cuff V2 Hose and orifice resistance System compliance Deflate R + Transducer P(0) Figure 5.1: Simplified electrical analog of pneumatic system, sensor. Initially, the cuff is inflated and hence the system compliance C is "charged" with a pressure P(0). Opening the deflation valve V"2 to the atmosphere for a known time T will produce a pressure decrease AP in the sensor. The resistance-compliance product, or time constant, of the pneumatic system can then be estimated as T In (P (0)) - In (P (0) - AP) (5.1) RC = Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 61 Following this simple calibration, the time td which V2 must be open to cause a step-deflation of Pdefl from any initial pressure Pinit is thus U= RC- (In (Pinit) - In (Pinit - Pdef,)) (52) In the context of oscillometric LOP estimation, a linear step-deflation algorithm would thus operate as follows. The sensor is first inflated to its maximum pressure, and this is followed by a calibration deflation to estimate the system time constant RC. During each subsequent data acquisition cycle, the deflation valve-on time td required to produce the desired decrease from the current sensor baseline pressure to the next is computed from (5.2) after the detection of each valid oscillometric pulse. Provided that Pdefl in (5.2) remains constant, stepwise-linear sensor deflation is produced by non-linearly increasing the deflation valve-on time td at each new baseline pressure. The above algorithm was implemented in the C language using 16-bit integer arithmetic and a 256-entry lookup table for the natural logarithm. The simple analog shown in Figure 5.3 is not completely descriptive and consequently the algorithm had to be slightly modified to produce perfectly linear deflation. Thus, if a pressure step was less than desired, the pneumatic system time constant estimated from the calibration step was increased by 5%, and if a pressure step was more than desired, the time constant was reduced by 5%. Figure 5.2 shows the sensor deflation curves obtained with the PVC cuffs on upper and lower 250 1 1 1 1 1 j 1 1 1 1 j 1 1 1 , j : ) 1 1 1 1 1 1 X s E. km 3 200 150 100 50 Constant valve-on time Controlled valve-on time 300 . 250 5 10 15 20 Deflation valve activations 50 Constant valve-on time Controlled valve-on time 5 10 15 20 Deflation valve activations 25 (a) (b) Figure 5.2: Step-deflation curves obtained using a constant deflation valve-on time and valve-on time controlled by the algorithm: (a) upper limb cuff on arm; (b) lower limb cuff on leg. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP limbs using a constant deflation valve-on time and using the controlled valve-on time algorithm. The horizontal axis is intended to represent the deflation valve activations caused by the detection of valid oscillometric pulses during a estimation cycle. The constant valve-on time deflation curves are exponential with pressure steps that decrease from 9 mm Hg to 3 mm Hg. The algorithm, however, produces the desired stepwise-linear deflation curves with nearly uniform 7.5 mm Hg steps. Note that the curves for the upper and lower limbs are equally stepwise-linear, indicating that the time constant approach employed is capable of compensating for the effects of cuff application, volume, and initial inflation pressure. 5.1.2 Deflation artifact inhibition Although the step-deflation technique provides a simple and effective method for sensor pressure control, each activation of the deflation valve excites the step response of the pneumatic system as illustrated in Figure 4.4. If unattenuated, this response will saturate the oscillometric signal amplifier resulting in a corruption of the pulse or pulses immediately following the deflation Fortunately, this problem may be corrected by improving the signal acquisition electronics with additional hardware as illustrated in Figure 5.3. The diagram shows that the wideband oscillometric amplifier in Figure 4.2 has been replaced by the near-optimum filter implementation described in Section 4.5.3, which is followed by an 8-step variable attenuator based on a C M O S analog multiplexer Instrumentation amplifier Pressure * transducer rx^\n^\ , Sensor baseline pressure Near-opt imum B P F 3.2 - 8.0 Hz 8 - s t e p attenuator To - analog multiplexer (Fig. 4.2) Figure 5.3: Improvements to signal data acquisition electronics. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 63 and an amplifier with a gain of 44. The attenuator receives a 3-bit number from the control logic block to regulate the amplitude of the oscillometric signal which is fed to the A - D converter. Also shown is an analog switch which shorts the first section of the Bessel filter to ground in response to a signal from the control logic. Activating the analog switch and setting the signal attenuation at the filter output to maximum before the deflation valve is opened prevents the step response of the pneumatic system from saturating the amplifier, which would otherwise delay the detection of oscillometric signals. Figure 5.4 illustrates the effectiveness of the analog switch and attenuator in eliminating the step-deflation artifact. A dual bladder P V C cuff was applied to a plexiglass cylinder and was connected to the I B M PC with a 3 m hose to obtain these results. The top trace of Figure 5.4a shows the amplifier output in response to a -10 mm Hg pressure step caused by opening the deflation valve to air for 30 msec, which is indicated by the pulse shown in the bottom trace. With the analog switch disabled and with minimum attenuation, the amplifier saturates and does not recover completely for nearly 900 msec. Figure 5.4b illustrates the amplifier output during deflation with the extra hardware in operation. By turning on the switch and selecting maximum attenuation 15 msec before the deflation valve is opened, and then turning off the switch and selecting minimum attenuation 155 msec after the valve closes, the artifact is completely eliminated. This technique therefore reduces the time during which i • 0.5 1 1.5 Time (sec) (a) 2.5 1 1.5 Time (sec) (b) Figure 5.4: Step-deflation artifact observed at amplifier output. Upper traces are the amplifier output signal and lower traces show the deflation valve control pulse, (a) Analog switch and attenuation disabled; (b) Analog switch and attenuation enabled. 64 ChapterS: Development and Evaluation ofan Algorithm for Estimating LOP pulses cannot be observed from 900 msec to 200 msec. 5.2 Detection Algorithm for Pulse Data Acquisition 5.2.1 Algorithm description The estimation of LOP may be divided into two distinct operations, the first being the acquisition of oscillometric pulses at decreasing sensor pressures, and the second being the off-line analysis of the pressure and pulse amplitude data to obtain the LOP. In a practical implementation, the rules for oscillometric pulse detection must operate in real-time on incoming data since the detection of pulses must be integrated with step-deflation control of the sensor. This real-time requirement inspired the development of an intuitively simple technique for oscillometric pulse detection. Heuristic, pattern-based schemes have been successfully employed in the detection of epileptiform spike discharges in the EEG [76, 77] in the location of the various waves and complexes in the ECG [78, 79], as well as in the diagnostic classification of ECG waveforms [79 - 81]. Similar to the patterns seen in the ECG, the signal variations observed during an oscillometric measurement also show distinctive patterns which may be used in the detection of valid pulses and the rejection of corrupted data. Figure 5.5 shows the sensor pressure (top trace) and the oscillometric pulses observed at the output of the near-optimum filter (bottom trace) as the sensor was deflated from supraocclusive to subocclusive pressures. In this case, the sensor was deflated continuously through a small orifice so "200 E E T 150 (a) 1 I 100 u o tf) 5 50 150 a a (b) | o 1/5 V JU 3 &. -150 0 2 4 6 8 10 12 14 16 Time (sec) Figure 5.5: Sensor signals observed during deflation: (a) Sensor baseline pressure; (b) Oscillometric pulse signal. 65 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP that the entire oscillometric pulse train could be observed without the discontinuities that would have been introduced by step-deflation. This illustration shows two distinctive features of oscillometric signal behavior during sensor deflation which can be used in the detection of pulses. First, pulses occur at regular, nearly deterministic intervals. Second, the variation in pulse amplitude and shape is quite small between adjacent pulses. A consequence of these patterns is that pulse amplitude, slope, and timing features can be used to define a heuristic detection rule as follows: i. Given that the peak-to-peak amplitude Ai and slope Si of a valid pulse i are known, Figure 5.5 shows that these values may be used to set highly restrictive amplitude and slope thresholds for pulse Hence, pulse i+1 is considered valid if, for i > 0, £>i = K\A{ < Ai+i < 1(2Ai = D2 (53) and D3 = K3Si < Si+1 < IuSt = D4 (5.4) where Kj to K4 are constants and Dj to D4 are the resulting adaptive detection thresholds. For the case i = 0, this rule presumes that the first pulse would be detected by comparing A0 and S0 with absolute amplitude and slope thresholds Amin,o> Amax,o> Smin.o and Smax.o-ii. After the two first pulses are detected, the time difference between them, Tj - To, can be used to set a minimum time after which pulse i+1 must occur, given that the time 7/ of pulse / is known. An upper bound on the time of pulse i+1 can also be specified in terms of the longest expected inter-systole interval, which is approximately two seconds [82]. These rules specify a window for the time of pulse i+1 as D5 = Ti + Ks {Ti - To) < r,-+1 < T, + 2 sec = D6 (55) Figure 5.6 is a schematic illustration of how these rules would operate in the detection of the third oscillometric pulse in a data acquisition cycle. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 66 time (D < A < D ) and (D„<S < D J and (D C <T <D ) v 1 2 2' v 3 2 A' 5 2 6' THIRD PULSE DETECTED Figure 5.6: Schematic illustrating detection of third oscillometric pulse. Corruption of the oscillometric pulses comes from two noise sources. The first is produced by the near-optimum filter itself in response to an input pulse, and appears as the low level fluctuations or ringing which follows the output pulse (Fig. 5.5). The second source of corruption is the remnant artifact which can appear at the filter output in response to surgical manipulation (Fig. 4.20, Fig. 4.22). The proposed detection rule would be effective in rejecting these corruptions since in the first case, the amplitude of the fluctuations is significandy less than that of the pulses, and in the second case, surgical artifact can have much greater amplitude than the pulses and is furthermore asynchronous with the heart systole. Figure 4.22 also illustrates that simple features such as positive or negative peak signal clipping can be used to identify and therefore reject many surgical artifacts. Rules which limit the maximum signal excursions are readily incorporated into the thresholds of the above pulse detection scheme. 5.2.2 Algorithm implementation Figure 5.7 is a block diagram of the real-time digital signal processing subroutines which were used in the practical implementation of the heuristic pulse detection algorithm. Each subroutine was implemented as a C language function using 16-bit integer arithmetic, with calculations optimized to satisfy the real-time processing requirement. The preprocessor algorithm, shown in Figure 5.8a, samples the oscillometric signal, and then searches for a local minimum followed by a local maximum which will define a rising segment in the Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 67 Signal segment amplitude, slope, and time Input data Signal preprocessor Amplitude and slope correction Pattern recognition Control data to analog attenuator Dynamic range compression Valid oscillometric pulse amplitudes Figure 5.7: Real-time signal processing algorithms for pulse detection, data. As indicated in the flowchart, the time 7/ of the rising segment is defined as the time of its minimum, its amplitude Aj is defined as the difference between the maximum and the minimum, and its slope Si is approximated by its amplitude divided by its duration. The use of the variable exit Jlag prevents the preprocessor from capturing noise spikes and notches of duration equal to three samples as shown in Figure 5.8b. These spike and notch artifacts may be introduced by drilling or chiseling as illustrated in Figures 4.19 and 4.20, or at much lower levels may be a natural characteristic of the oscillometric pulse morphology of the patient. The signal features A/ and Si are then scaled and combined with 7/ to become the input parameters for the detection rules given by equations (5.3) - (5.5). If these quantities fall within the detection thresholds Dj through D(>, the segment is declared a valid pulse and its amplitude Ai is stored along with the cuff baseline pressure at which it was observed. D ; to D<5 are then updated according to the rules (5.3) - (5.5) in anticipation of the detection of pulse A procedure which operates on the A,- data at the output of the detection algorithm is a dynamic range compressor. If a detected oscillometric pulse amplitude as referenced to the input of the A - D converter exceeds 128, a control code is sent to the 8-step variable attenuator shown in Figure 5.3 to increase the analog signal attenuation by one step. Conversely, if the pulse amplitude at the A - D converter input is less than 90, the compression algorithm decreases the analog signal attenuation by one step. The compression algorithm thus acts to minimize quantization distortion while preserving 6 dB of converter headroom. This headroom reduces the likelihood of signal clipping from unanticipated pulse amplitude increases or from motion artifact. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 68 min = 128 ex iM lag = 0 get data k = k+1 min = data Tmin = k ex iM lag = ex iM lag + 1 ex iM lag = 0 Y N max = -128 ex iM lag = 0 get data k = k+1 max = data Tmax = k e x i M l a g = ex iM lag + 1 ex iM lag = 0 Y N | pattern 1 recognition Tj = Tmin = max - mm Ai Sj = Aj /(Tmax - Tmin) (a) 69 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP A s shown in Figure 5.7, the segment amplitude and slope data at the preprocessor output are scaled to correct for the changes in acquisition conversion gain introduced by the compression. Segment amplitudes and slopes, which without the compression algorithm could have been represented as 8-bit quantities, are thus converted to 16-bit integers. Figure 5.9 is a detailed flowchart of the pulse acquisition algorithm, illustrating the integration of the preprocessor, detection rules, and dynamic range compressor with the sensor step-deflation control procedure developed in Section 5.1. In addition to (5.3) - (5.5), other rules have been included in this practical implementation to deal with foreseeable special cases. These rules appear in the flowchart in Figure 5.9, and are described below. a. After location of a rising segment, the local maximum and min imum are tested and rejected if the signal has been cl ipped due to large motion artifacts. b. If there is no converter saturation from motion artifact and the first pulse is not detected within two seconds, the sensor is step-deflated to increase the signal amplitude. c. If pulse i > 0, is not detected within its specified time window D($ - Ds, the acquisition algorithm is restarted. The assumption behind this rule is that i f pulse i+1 is not located, large motion artifacts must be present which would corrupt the amplitude of the pulse and perhaps later pulses as wel l , thereby resulting in an erroneous estimate of L O P . Al though it might seem that restarting the entire acquisition cycle due to high level artifact would greatly extend the time between measurements, it w i l l be shown later that an uncorrupted sequence of only a few pulses is actually needed to estimate the L O P , and hence this rule is appropriate. d . In order that the inter-pulse interval Tj - To be correcdy determined, accurate detection of the first pulse is mandatory. Therefore, as shown in Figure 5.9b, the amplitude A0~ and slope S0-of the fal l ing edge of the first pulse are extracted with a second preprocessor and compared to thresholds after the rising edge has been detected. Using a similar approach as that for later pulses, the thresholds for the fall ing edge of the first pulse are obtained from the features of the rising edge. e. The sensor of a leg tourniquet has a particularly large volume, and hence the time td required to open the deflation valve to produce the desired pressure drop may be up to several hundred mill iseconds at lower sensor pressures. Deflation of the sensor may therefore interfere with pulse detection. Hence, Figure 5.9b shows that i f the deflation valve-on time td extends ChapterS: Development and Evaluation of an Algorithm for Estimating LOP Start inflate sensor to maximum pressure D 1 - A M m i n , o D 2 = A m a x , o D 3 = Smin .o D 4 = c °max ,o D 5 = 0 D 6 = 2 s e c i = 0 time of pulse i amplitude of pulse i slope of pulse i detection threshold constant data acquisition and preprocessor r (Fig. 5.8) amplitude and slope correction Y D 5 = T 0 D 6 = T o + 2 sec step-def late sensor CD store A j , S j , T j F igu re 5.9: (a) Real-t ime pulse detection and sensor deflation control algorithm. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 9 get next data N preprocessor 2 negative edge A 0 ' . S o " D 5 = T 0 D 6 = T 0 + 2 sec N analog switch on T j = time of pulse i A j = amplitude of pulse i S j = slope of pulse i Dj = detection threshold K : = constant correct time constant calculate t d (Eqn. 5.2) open valve for t d sec delay 160 msec analog switch off D 5 = • T i + K 5 ( T 1 • V D 6 = • T i + 2 sec (Eqn. 5.5) D 5 - D 5 + " T 0 ) D 6 = D 6 + " T 0 ) D 1 . K l A j D 2 = K 2Aj (Eqn. 5.3) D 3 = K g S j D 4 = K 4 S j (Eqn. 5.4) F igu re 5.9: (b) Real-time pulse detection and sensor deflation control algorithm (continued). 72 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP beyond the lower bound Ds o f the time window for pulse i+1, the window is delayed by one inter-pulse interval Tj - To. In this case, the algorithm w i l l search for the second rather than the first pulse fol lowing the sensor step-deflation. The performance of the algorithm in extracting pulses from corrupted data is mostly determined by the selection of the detection constants K] to Ks in (5.3) - (5.5), and to a lesser extent is affected by the choice of the first pulse detection thresholds Amin,o, Amax,o, Smin,o and Smax.o- Since the algorithm wi l l decrease the sensor pressure until it finds a signal, the specification of Amini0 and Smin,o is somewhat arbitrary. Consequently, Amin,o was chosen to be one-tenth o f the A - D converter dynamic range and Smin,o was heuristically specified as Amin,o divided by 5 time samples. The compression algorithm also acts to l imit the signal to one-half the dynamic range of the A - D converter, and hence Amax.0 was heuristically chosen as one-quarter o f the converter dynamic range and Smax,o was set to Amaxt0l3. Observations such as those illustrated in Figure 5.5 can be used to empirical ly select Kj through Ks to achieve a desired pulse detection or noise rejection performance in the laboratory. The I B M P C data acquisit ion system was modif ied to include the hardware improvements shown in Figure 5.3, and the algorithm of Figure 5.9 was used to acquire pulse time, amplitude, and slope data f rom the upper and lower l imbs of male and female volunteers under noiseless conditions. Initially, the detection constants were arbitrarily specified as Kj = K3 = K5 = 0.4 and K2 = K4 = 2.5. Inspection of the acquired pulse data and further empirical tuning of the parameters in the laboratory during conditions of l imb manipulation produced the values Kj = 0.63, Kj = 2.0, K3 - 0.50, K4 = 2.0 and Ks = 0.6. These choices greatiy improved the qualitative performance of the algorithm in terms of both pulse detection and motion artifact rejection. 5.2.3 Per fo rmance Eva lua t ion The performance of the pulse detection algorithm was quantitatively evaluated using the signal data obtained from the total knee arthroplasty performed by Dr. R. Claridge as described in Section 4.5.3. A total of 37 8-second signal samples were obtained during this operation, each sample being an observation of oscil lometric pulses under different artifactual conditions as illustrated by Figures 4.18 to 4.22. A n off-line version of the algorithm shown in Figure 5.9 was implemented without the sensor deflation and dynamic range control procedures, and the detection constants Kj to Ks were set to the empir ical ly determined values discussed in Section 5.2.2. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 73 Figures 5.10a through 5.10d are examples illustrating the detection performance of the algorithm for signals corrupted by different artifacts. In Figure 5.10a, the signal was corrupted by dr i l l ing the femur. The signal data appears in the upper trace and the lower trace indicates the time at which the algorithm detected a valid pulse. This figure shows that the algorithm is effective in rejecting the low level sharp artifacts which are characteristic o f this type of surgical manipulation. (a) (b) (c) (d) J l II II IL J IL L_i L. I I _l I I I i i i A A A A A A A A A 3 4 Time (sec) F i g u r e 5.10: Performance of the pulse detection algorithm on signals corrupted by various surgical artifacts. Upper traces are the oscil lometric signal and lower traces indicate detections declared by the algorithm, (a) Dr i l l ing artifact; (b) hammering artifact; (c) moderate motion artifact; (d) extreme motion artifact (arbitrary y scales). 74 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP The other extreme is represented in Figure 5.10b, in which large spike artifacts are present from hammering on the femoral component. In this case, the algorithm rejects the artifacts since their amplitudes and slopes are greater than the maximum permissible amplitude and slope thresholds defined by the previous pulses. Figures 5.10c and 5.10d are examples illustrating the response of the algorithm to motion artifact. In Figure 5.10c, the filter suppresses most of the artifact and the pulse detection algorithm makes no errors. When motion artifact becomes extreme, Figure 5.10d shows that the algorithm declares two false positives, although it is capable of correctly extracting two of the three valid pulses which appear between 2 and 4 seconds. O f the 37 signal samples, 14 could be classif ied as examples of high level mot ion artifact as illustrated in Figure 5.10d. A graphical output f rom the off-line detection algorithm l ike that shown in Figure 5.10 was used to visual ly verify correct, false positive, and false negative pulse detections. Out of a total o f 253 visual ly recognizable pulses, 209 were correctly identified to produce a correct detection performance of 83%. In 59 cases, the algorithm detected pulses which were subjectively determined by visual inspection to be artifacts. Hence, the false positive error was 59/253 or 23%. The algorithm missed 253 - 209 = 44 val id pulses, producing a false negative error of 17%. Remembering that about 40% of the signal samples used to obtain these results were acquired under conditions of high level artifact, the algorithm is apparently very good, although not excellent, at extracting pulses in noise. This sensitivity, however, is reflected in the false positive error, and it was noted that almost all false positives occurred during highly artifactual conditions such as that il lustrated in Figure 5.10d. Poor synchronization to the heart rate, which arises from an error in the detection of the first or the second pulse but not both, w i l l produce large increases in the false negative error since all fo l lowing pulses are l ikely to be missed. O f the six types of orthopaedic procedures which were observed in later cl in ical trials of the adaptive tourniquet system, none generated greater surgical artifact than a total knee arthroplasty. E v e n so, the total amount of time in which high level artifact l ike that in Figure 5. lOd occurs is at most one or two minutes out of a total operating time of typically two hours. The amount of time that a bone saw, a dr i l l , or a osteotome contacts the l imb to produce artifacts l ike those illustrated by Figures 5.10a or 5.10b is also insignificant when compared to the length of the procedure. The most frequently occurr ing artifact is produced when the l imb is manipulated slightly as instrumentation or prosthetic sizers are first brought in contact with or removed from the l imb, and this condition is represented by 75 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP Figure 5.10c. Since it is anticipated that the periods during which artifact is extreme w i l l be relatively short, appropriate integration of the algorithm into a practical system with additional features to deal with this situation may stil l result in effective overall performance. This point is validated by the results of the cl inical trials which are presented in the fol lowing chapter. 5.3 Pu lse Ana lys is A l g o r i t h m for L O P Es t ima t ion The auscultatory estimation of central arterial systolic blood pressure (SBP) actually involves the measurement of the L O P of a l imb-occluding cuff. The auscultatory and oscil lometric techniques of S B P estimation are known to correlate wel l [40], and both methods are known to produce falsely elevated S B P ' s i f the cuff width becomes less than 40% of the l imb circumference [5, 44]. A s discussed in Chapter 3, this is because the L O P of the occluding cuff increases with decreasing cuff width. These facts suggest that since the results obtained by auscultatory and oscil lometric means are simi lar for specific width-to-circumference ratios of cuffs, the rules used in osci l lometric S B P estimation may be directly applied to cuff L O P estimation and vice versa for the same ratios. This is an assumption which forms the basis of the development of the L O P estimation method presented here, and which was later validated by the results obtained in an experiment which evaluated the accuracy of the technique. The oscil lometric estimation of blood pressure relies on the observation of two phenomena. Yelderman and Ream showed that, during the deflation of a blood pressure cuff of width between 40 and 60 per-cent of the l imb circumference, the cuff pressure of maximum oscil lometric pulse amplitude was equal to the mean arterial pressure ( M A P ) [41]. B y simultaneously recording osci l lometr ic data and Korotkov sounds while deflating a blood pressure cuff, Geddes found that the osci l lometric pulse which occurred synchronously with the onset of f low as indicated by the first Korotkov sound had, on average, an amplitude of 55% of the maximum observed amplitude [42]. These phenomena, hereafter referred to as the MAP rule and the Geddes rule respectively, form the basis of the M A P and S B P estimation methods used in many commercial oscil lometric b lood pressure monitors [27]. There would be three disadvantages to the use of the Geddes rule assuming that it could be directly applied to the estimation of cuff L O P . First, the cuff would have to be deflated from above L O P to below M A P to obtain the pulse amplitude at L O P and the maximum pulse amplitude. It w i l l be shown later that the time required to perform this wide pressure sweep and acquire the necessary data 76 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP with a conventional algorithm is well beyond the 10-second inter-estimate interval specified at the beginning of this chapter. Second, when the cuff pressure was above LOP, the oscillometric signal would have an amplitude less than one-half its maximum value, and hence the pulses would be more easily corrupted by artifacts. Finally, reports of patient injuries from blood pressure monitors [83, 84] indicate that the repetitive cycling of the cuff pressure over such a wide range could in fact exacerbate the neural injury problem the adaptive tourniquet would try to prevent In light of these facts, an alternative method of estimating LOP was sought using a computer model of the cuff-limb system developed by Forster and Turney [85]. An examination of their simulation results inspired the idea that the cuff-limb system became linear when the cuff pressure was equal to the MAP, and hence changes in the arterial pulse pressure would be reflected as proportional changes in the maximum oscillometric pulse amplitude. This original idea is examined in the following section using the Forster-Turney model, and the results from the computer simulation are used to develop a new method for estimating LOP. 5.3.1 Computer simulation of oscillometry 5.3.1.1 Model and system equations Forster and Turney assumed that the cuff-limb system could be represented as the one-dimensional model shown in Figure 5.11a. The model, which is lossless and does not incorporate inertial effects, assumes that the limb tissue is incompressible and the outside surface of the cuff is rigid. The model is assumed to be axially uniform and of an arbitrary length so that the system equations which describe the model may be written in terms of the arterial and cuff volumes. Although the model is used here to develop rules for SBP estimation, the results should be applicable to LOP estimation based on the assumption discussed previously. An electrical analog of the biomechanical model is shown in Figure 5.1 lb, in which the arterial blood pressure pa(l) excites two series-connected, time-varying compliances which describe the volume-pressure (v-p) characteristics of the artery and cuff bladder. The oscillometric signal is the a.c. component of the pressure pc(t) observed across the cuff compliance, which at t = 0 is initially "charged" with a pressure Pco. The v-p characteristic of the cuff is described by the ideal gas law Pc(t)vc(t)k = PCOVCO k (5.6) 77 ChapterS: Development and Evaluation of an Algorithm for Estimating LOP arterial lumen limb tissue arterial wall (a) (b) F igu re 5.11: (a) Biomechanical model o f cuff- l imb system; (b) electrical analog of model, where vc(t) is the cuff volume and VCo is the volume at t = 0. The gas constant, k, is 1.4 for air. The v-p characteristic o f the artery is known from in vitro studies to be exponential [86]. Forster and Turney showed that, for non-pathological arterial compliance, the v-p characteristic which produces the most realistic simulation results is given by va (t) = 12.3 ( l - e-°' 0 2 2P ' ( ( ) j + 3.3 mm 3 , pt (t) > 0 (5.7a) and va(t) = 3.3e°-082>"W mm 3, pt (t) < 0 ( 5 / 7 b ) where va(t) is the arterial volume and pt(t) is the transmural pressure across the arterial wal l , defined as Pt(t) = P a (t)-Pc(t) (5.8) 78 ChapterS: Development and Evaluation of an Algorithm for Estimating LOP The arterial pressure, pa(t), was approximated in the original study as a half-wave rectified sinusoid wi th a duty cycle of 42%. Wri t ten as a discrete time series of length 100 samples, the arterial pressure wave is thus p . ( » ) = { P > s i n ( ^ ) + D B P 0 < n < 41 42 < n < 100 (5.9) where DBP is the diastolic blood pressure and Pp is the arterial pulse pressure, defined as Pv = SBP - DBP. (5.10) The M A P may be obtained f rom the time average va(t) + vc(t) = VT (5.11) for comparison with the osci l lometric M A P estimates. Since the outside of the cuff is assumed rigid and the l imb tissue is incompressible, the total system volume, Vr, must remain constant. Thus M 99 (5.12) n=0 where a realistic value for Vj would be approximately 100 times the arterial volume at zero transmural pressure [85]. Combining (5.6) - (5.12) produces the two nonlinear equations 2 . 3 ( l - e - ° - ° 2 2 ( P . ( n ) - P c ( n ) ) ^ ) + 33 + Pc{n) V T , Pa(n)>Pc{n) (5.13a) and 3 3 e0.0S2(po(n)-pc(n)) + .Pc(n) = VT, Pa{n)<pc(n) (5.13b) wh ich must be solved numerically for pc(n) at each discrete value of pa(n) given an initial cuff pressure Pco. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP The init ial cuff volume can be obtained from (5.12) as Vo — Vp — va (0) (5.14) where va(0) is calculated from (5.7a) or (5.7b) withpt(0) = DBP - Pco. Once pc(n) has been obtained, its a.c. component can be extracted and the peak-to-peak amplitude calculated to produce the osci l lometr ic pulse amplitude observed at cuff pressure Pco. 5.3.1.2 Simulation experiments and results T w o simulations were written in the C language to observe the effects of cardiovascular dynamics on the osci l lometric signal. In both simulation experiments, the cuff pressure PCo was initialized at 170 mm H g and was decreased in 3 mm Hg steps to a min imum pressure of 53 mm Hg . A t each baseline pressure the amplitude of the oscil lometric signal was calculated and stored along with its corresponding cuff pressure. The cuff deflation cycle was repeated 13 times, and between each cycle the arterial pressure wave was altered to simulate a changing cardiovascular variable. The first simulation was intended to show the effect o f a change in the total peripheral resistance (TPR) of the vasculature on the oscil lometric signal. Autonomic control o f vasoconstriction or vasodilation w i l l result in an equal shift in the S B P and the D B P [82], and hence in this simulation, Pp was kept constant at 40 mm H g while SBP was increased in 5 mm H g steps between each cuff deflation cycle from 60 mm H g to 120 mm Hg . Figure 5.12 shows the oscil lometric amplitude obtained as a function of cuff pressure for the cases SBP = 100, 130, and 160 mm Hg . The results show that the peak of the amplitude distribution shifts horizontally right to reflect the increase in M A P , which is accompanied by a 60% increase in the peak signal amplitude. The second simulation was intended to show the effect of an increase in cardiac stroke volume (SV) on the oscil lometric data. According to Starling's Law [82], S V increases due to an increase in venous return to the right heart, which may result from a positive change in cardiac output due to an increased heart rate. The effect of increased S V at constant T P R is to increase the arterial pulse pressure, Pp [82]. This was simulated by varying SBP from 100 to 160 mm Hg in 5 mm Hg steps with DBP = 80 mm Hg . This produced the results shown in Figure 5.13 for SBP = 100, 130, and 160 mm H g . In this case, the horizontal shift in the peak of the distribution is not as pronounced, although a much larger increase of 260% in the peak signal amplitude is observed. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 80 oo X E E a "a. E < o •c 4—1 E o 25 20 15 10 Systolic = 100 mm Hg -Systolic =130 mm Hg -Systolic =160 mm Hg -50 90 110 130 Cuff Pressure (mm Hg) 170 Figure 5.12: Osci l lometr ic pulse amplitudes obtained while varying total peripheral resistance. 00 X E e, <u T3 3 "S. E < o "S E o a 25 I I I I I 20 15 10 i i i i | i i i i | i Systolic = 100 mm Hg Systolic = 130 mm Hg Systolic =160 mm Hg 50 70 90 110 130 150 Cuff Pressure (mm Hg) 170 Figure 5.13: Osci l lometr ic pulse amplitudes obtained while varying cardiac stroke volume. Figure 5.14 shows the linear relationship obtained from the second simulation between the peak signal amplitudes Ap and the pressure difference Pd = SBP - DBP = Pp. L inear regression [88] was used to obtain the line drawn through the simulation data. The correlation coefficient Rxy was calculated to be in excess of 0.999, indicating that a change in pulse pressure wi l l be directly reflected 81 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP as a proportional change in the maximum pulse amplitude. A similar highly linear relationship was also found between Ap and Pd = SBP - MAP. Pd, Arterial Pressure Difference (mm Hg) F i g u r e 5.14: L inear relationships between maximum osci l lometric pulse amplitude Ap and Pd = Pp, and Ap and Pd = SBP - MAP. It is evident from Figure 5.12 that variations in S B P due to changes in T P R may be tracked by observing the cuff pressure of maximum signal amplitude and adding a constant pressure offset to this value to estimate the S B P . Wi th changes in S V , however, Figure 5.13 shows the cuff pressure of max imum signal amplitude does not track the S B P with a constant pressure offset. The linear relationships shown in Figure 5.14 suggest that the variable offset between S B P and M A P which results from changes in S V may be calculated from observed proportional changes in the peak pulse amplitude. S B P could thus be estimated by f inding the cuff pressure of maximum signal amplitude and then adding an amplitude-compensated pressure offset to this value. The fol lowing section compares the S B P estimation accuracy of this approach to that obtained using the Geddes rule. 5.3.1.3 Comparison of SBP estimation methods A s shown by Forster and Turney, the M A P rule was found to be an accurate, consistent estimator of the true M A P . After examining the results of all simulations, the largest difference between the cuff pressure of max imum oscillometric amplitude and the M A P was +4.9 mm Hg. 82 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP Us ing the data from both simulation experiments, three techniques for estimating S B P were compared on the basis of the difference between the known arterial S B P and the estimated S B P , defined as the S B P estimation error. The first technique evaluated was the Geddes rule, which produced estimates SBPej. In the second technique, the Geddes and M A P rules were used to init ially define a pressure offset P0 from the S B P and M A P estimated at an arterial systolic pressure of 120 mm H g . Thus P0 = SBPeA -MAP, (5.15) where MAPe is the M A P estimate obtained by applying the M A P rule to the simulation data. A l l subsequent S B P ' s were then estimated by using the rule given by SBPe<2 = MAPe + P0 { 5 U ) In the third technique, the pressure offset P0 was compensated using the amplitude Ap of the osci l lometr ic pulse observed at the cuff pressure equal to MAPe. Thus 5 5 P e , 3 = M APe + VLZM (5>17) where A p o is the amplitude of the pulse observed at the cuff pressure equal to the MAPe used in (5.15) to calculate P0. Figure 5.15a shows the estimation errors as a function of S B P using the three techniques on the osci l lometr ic data from the first simulation. The Geddes rule produces the lowest estimation errors in this case, with a peak error o f -3 mm H g occurring at S B P = 140 mm Hg . The third technique produces an error of -4 mm Hg at low S B P and an error of +4 mm H g high S B P . This suggests that the amplitude variation observed with changes in T P R slightly overcompensates the offset pressure PQ. However , as all three techniques produce errors which are less than +/- 5 mm H g , their accuracies are comparable and hence all may be considered equally responsive to changes in T P R . Figure 5.15b shows the estimation errors obtained using the three techniques on the data from the second simulation. A s was predicted by Forster and Turney, the Geddes rule produces low estimates at high pulse pressures. The second technique, in which a constant offset is added to the M A P estimate, Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP produces extremely large errors in response to variations in stroke volume. The third technique, which compensates the offset based on the observed maximum pulse amplitude, produces the lowest errors of al l three methods. -Si X E E W G E 20 10 T—i—i—i—|—i—r—i—i—|—i—i—r Geddes Rule O--MAP esti mate + offset — B -— MAP estimate + corrected offset — A -•10 ~ -20 C/3 -30 -a x E E 20 10 - i — i — i — r T Geddes Rule O---MAP estimate + offset — B — -"N><taMAP estimate + corrected offset — A — 100 120 140 160 Arterial Systolic Pressure (mm Hg) (a) 100 120 140 160 Arterial Systolic Pressure (mm Hg) (b) F i g u r e 5.15: (a) S B P estimation errors obtained with oscil lometric data from the first simulation; (b) estimation errors obtained with oscillometric data from the second simulation. Th is last method has other practical advantages over the Geddes rule algorithm besides improved accuracy when applied to L O P estimation. First, the cuff can be swept over a much narrower range of pressures to locate the peak signal, thereby greatly reducing the data acquisition time. Second, the pulse amplitude at these cuff pressures is near maximum, as is the signal-to-noise ratio. F ina l ly , the cuff pressure remains at low subocclusive values near 100 mm H g , and hence the risk o f cuff-induced neural injuries is negligible. However, since the method uses cuff pressures which are subocclusive, the technique is only suitable for use in a dual bladder adaptive tourniquet system where a distal occlusive bladder would prevent venous congestion. 5.3.2 A l g o r i t h m implementat ion The compensated-offset method of oscil lometric S B P measurement was implemented as a C language program for estimating the L O P of the tourniquet cuff sensor using the I B M P C data acquisit ion hardware. The program, which incorporated the pulse detection algorithm developed in Sect ion 5.2, obtained the L O P as illustrated in the flowchart in Figure 5.16. er 5: Development and Evaluation of an Algorithm for Estimating LOP ( j a r ? ) inflate sensor to supraocclusive pressure I pulse detection and sensor step-deflation algorithm (Fig. 5.9a, b) sensor pressure = 60 mm Hg Y Geddes Rule LOP . , MAP , A e,1 e p,o o = L O P e , 1 " M A P e < E q n - 5 - 1 5 ) inflate sensor to M A P e + 2 0 m m H g pulse detection and sensor step-deflation algorithm (Fig. 5.9a, b) M A P Rule M A P , V ( A P . o + V V i - ^ 3 I (Eqn. 5.18) L O P r M A P e + ( P o A ^ / A p i 0 (Eqn. 5.17) (Eqn. 5.19) L O P e =(LOPj+ L O P M + L O P j . 2 ) / 3 L O P e = LOP j F igure 5.16: L O P estimation algorithm. 85 Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP The diagram shows that the first L O P and the constants P0 and Ap>0 are obtained from the Geddes rule. Fo l lowing this, the cuf f pressure is swept around the pressure of maximum pulse amplitude to obtain the subsequent L O P ' s . The pressure step size chosen for the sensor step-deflation algorithm requires a compromise between L O P estimation speed and estimate resolution. From practical experience with the estimation algorithm and laboratory volunteers, a compromise of -7.5mm Hg/step-deflation was determined wh ich was suitable for both upper and lower l imb tourniquet sensors. A s was done in previous adaptive tourniquet systems [5, 6], the L O P estimates were smoothed with a moving average to improve the estimate resolution and stabilize the output L O P values. Smoothing was applied to the data in two operations. First, the peak signal amplitude Ap used in (5.17) to compensate the offset P0 was obtained from a three-point moving average as Ap = (aP,Q + APJ + AP.i-i) ( 5 . 1 8 ) 3 where Ap>0 is the peak pulse amplitude observed from the Geddes rule L O P estimation cycle, Apj is the peak pulse amplitude observed in the current estimation cycle, and Apj.j is the peak pulse amplitude observed in the previous estimation cycle. The current occlusion pressure LOPi was calculated using (5.17), and the occlusion pressure estimate LOPe was then obtained from another three-point moving average as rnP - (LOPi + LOP^+ LOPj-2) e - 3 (5.19) In order to track rapid increases in the L O P due to changes in l imb position or blood pressure, smoothing was not applied to the estimates as given by (5.19) i f LOPi > LOPe + 15 mm Hg. In this case, the unsmoothed data point LOPi w a s u s e < 3 a s the L O P estimate. These averaging calculations are shown in the flowchart of Figure 5.16. 5.3 J Per formance evaluat ion 5.3.3.1 LOP estimation time From the flowchart in Figure 5.16, it is possible to calculate the improvement in L O P estimation time of the experimental system using the algorithm over that which would be obtained with an automated oscillometric blood pressure monitor such as the Cr i t ikon Dinamap 845, which uses an 86 Chapter 5: Development and Evaluation ofan Algorithm for Estimating LOP approach similar to the Geddes rule [39, 67]. It w i l l be assumed here that the first L O P estimation has been completed by both systems, and both are operating under non-artifactual conditions. For convenience, it is assumed that both systems are attempting to estimate an S B P of 120 mm H g with a blood pressure cuff of appropriate width, and that the D B P is 80 m m H g . The M A P , which is approximated here with a famil iar c l in ical rule [82], is (120 - 80)/3 + 80 = 93 mm H g . The Dinamap data acquisition algorithm is described in the service manual o f the instrument [67], and operates as fol lows. The cuff is first inflated to 40 mm H g above the last S B P estimate, or 160 mm H g . A t each baseline pressure, two pulses are detected, and i f they are matched, the cuff is step-deflated by 8 mm Hg . When the Dinamap has detected from the pulse amplitude data that the D B P has been reached at 80 mm H g , the S B P estimation rules are applied to the data to obtain the new estimate. The data acquisition time is therefore 2(160 - 80)/8 = 20 pulses or heartbeats. A t an average heart rate o f 72 bpm, this corresponds to 16.7 seconds. The experimental system would start by inflating the cuff to 93 + 20 = 113 mm H g , and after detection of a pulse, would step-deflate the cuff by 7.5 mm H g until the first pressure less than 93 - 20 = 73 mm H g was reached. The M A P rule would be applied and the new L O P estimate obtained from the M A P estimate plus the amplitude-compensated offset pressure. The data acquisition time is thus 45/7.5 = 6 heartbeats, plus one heartbeat for the inter-pulse interval determination performed by the pulse detection algorithm. A t a heart rate of 72 bpm, this corresponds to 5.8 seconds, about one-third the time required by the Dinamap 845. 5.3.3.2 LOP estimation accuracy The accuracy of the L O P estimation algorithm was tested in the laboratory on the upper l imbs of 8 male and female subjects using the adaptive tourniquet system described in the fol lowing chapter. The cuf f width-to-limb circumference ratio ranged from 0.19 to 0.34 for these subjects. The true L O P of the tourniquet was measured by deflating the distal bladder of the cuff while monitoring the radial pulse with an ultrasonic Doppler flowmeter. Since the tourniquet cuff sensor acquires the oscillometric data at subocclusive pressures, it is impossible to obtain paired estimates from the algorithm and the Doppler flowmeter. The fol lowing experiment was thus performed under the assumption that the L O P should remain constant during measurements made with both techniques provided that the entire trial was executed over a short time interval. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP The distal bladder of the tourniquet cuff was first inflated to a supraocclusive pressure and then s lowly deflated unti l the radial pulse was detected with the flowmeter, at which point the cuff pressure was recorded. The tourniquet pressure was then increased unti l the f low was again arrested, and this cuf f pressure was also noted. The true L O P was assumed to be the average value of these two tourniquet pressures. The distal bladder of the tourniquet was then inflated to a pressure which occluded the arm, and the L O P estimation algorithm was started. The adaptive tourniquet system was used to obtain six to ten L O P estimates using the sensor of the tourniquet cuff, which required approximately one minute or less. The estimation algorithm was then stopped, and the distal bladder o f the tourniquet deflated again to determine the L O P at the end o f the trial with the Doppler f lowmeter. Figure 5.17 shows a typical result obtained from the arm of a 24 year-old female. 130 i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 i ^ i 5 100 -Cu O Doppler LOP o -j 85 - Estimated LOP e—-70 1 L ; 1 1 1 1 1 1 1 1 1 1 1 1 1 0.5 1.0 1.5 2.0 Time (minutes) F igu re 5.17: L O P ' s obtained with ultrasonic Doppler f lowmeter and with estimation algorithm. The average o f the two L O P ' s measured with the flowmeter at the beginning and end of the trial was obtained and the differences between this value and the estimated L O P ' s were calculated. From a total of 16 trials consisting of 64 Doppler L O P measurements and 103 L O P estimates, the mean L O P estimation error was +3.0 mm Hg with a standard deviation of 8.1 mm Hg . These results are comparable to S B P estimation errors reported for the Cr i t ikon Dinamap 1848, an improved version of the Dinamap 845, which showed mean estimation errors of +6.3 mm H g with a standard deviation of 7.0 m m H g when compared to the results obtained by auscultation [89]. If the results obtained with the Dinamap 1848 can be considered a benchmark for typical osci l lometr ic S B P - L O P estimation accuracy, the results acquired here illustrate two points. First, as was found in Chapter 3, the improved tourniquet cuff must produce nearly the same surface pressure underneath both the proximal and distal bladders, otherwise the Doppler-measured L O P of the distal bladder would not correlate wel l with the estimated L O P o f the proximal bladder. Second, the initial Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP 88 assumption that osci l lometric rules can be directly applied to the L O P estimation problem has been validated here for cuff width-to-l imb circumference ratios ranging from 0.19 to 0.34. A second experiment was conducted to see i f the algorithm would produce the same L O P variations in response to l imb manipulation as those reported by Bussani [6, 7]. Figure 5.18 shows the L O P variations observed from the lower l imb of a 27 year-old male while the l imb was moved to simulate the surgical manipulations used in arthroscopy of the knee. The distal bladder o f the tourniquet was set to 260 m m H g to occlude the l imb in this experiment. The data shows similar L O P variations as those observed by Bussani , which are due in part to the elevation of the l imb with respect to the heart, and in part to changes in the diameter of the femoral artery with changes in l imb f lexion [87]. The l imb volume and geometry also vary with changes in l imb position, so that some L O P deviations may arise from variations in cuf f snugness. _ 230 60 X E a, c o 210 190 u O 170 E J B 1 5 0 E W 130 Limb Limb i i I i . , , I Limb 1 1 1 1 1 Level Level Level \ -Limb Limb Figure — - Elevated Fully 4 - Flexed Position " . . i l l i i I 1 i 0.0 2.0 4.0 6.0 8.0 Time (minutes) 10.0 12.0 F igu re 5.18: L O P variations observed during simulated arthroscopy of the knee. The upper l imb of the same subject was used to obtain the L O P data in Figure 5.19, which shows the estimation algorithm's response to two Valsalva maneuvers (t = 2.3 and t = 3.5 minutes). The voluntary increase in thoracic pressure produces a significant increase in the L O P estimate immediately fol lowing each maneuver, fol lowed by a gradual decrease in the L O P to the pre-Valsalva state as the subject relaxes. This response is a result o f the averaging which is applied to the L O P data, wh ich does not smooth the next estimate i f it is 15 mm H g greater than the running average, but which w i l l smooth all decreasing occlusion pressure estimates. Chapter 5: Development and Evaluation of an Algorithm for Estimating LOP _ 160 X § 150 u | 140 u a, § 130 J 3 73 120 I 110 •o g 100 E 90 ' 1 I 1 1 1 1 I ' ' ' ' I I I I I I I I . I I I I I T-Valsalva # 1 Valsalva # 2 1.0 1 1 1 1 1 1 1 1 1 1 1 1 ' 1 ' ' 1 ' < ' • I ' i ' ' I 1.5 2.0 2.5 3.0 Time (minutes) 3.5 4.0 4.5 Figure 5.19: Upper limb LOP variations observed in response to Valsalva maneuvers. 5.4 Summary Unsatisfactory results were previously obtained in clinical trials in which an oscillometric blood pressure monitor was used to estimate LOP for the adaptive control of tourniquet pressure. This inspired the development of an improved LOP estimation algorithm which was faster and had specific methodologies for coping with surgically related signal artifacts. A method for deflating the sensor bladder of the tourniquet cuff was implemented which could produce the same deflation characteristic independent of cuff application, volume, and initial pressure. The method is based on a model which assumes that the pneumatic system may be represented by a single resistance-compliance time constant. Observation of the pulse features observed during sensor deflation facilitated the development of a heuristic pulse pattern recognition algorithm in which the amplitude, slope, and time of a pulse are used to set detection thresholds for the following pulse. Evaluation of the detection algorithm on corrupted signal data obtained during a total knee arthroplasty showed a correct detection performance of 83%. However, the algorithm does declare artifacts as valid pulses in some cases when the noise level is high. Furthermore, long sequences of valid pulses can be missed if the first pulse is not accurately detected. These are the primary limitations of the approach, which should not create serious problems clinically if the integration of the algorithm into a system includes additional features to deal with these difficulties during the relatively brief intervals when they could occur. 90 ChapterS: Development and Evaluation of an Algorithm for Estimating LOP A new method for estimating LOP was developed from a computer simulation modeling the biomechanical interaction of the tourniquet cuff sensor bladder and the limb. The method is based on the addition of a pressure offset, which is a function of the maximum oscillometric pulse amplitude, to the pressure obtained by application of the MAP rule to the pulse data. Simulation results showed that the SBP estimation accuracy of this method is comparable to that achieved with the Geddes rule with changes in total peripheral resistance and is more accurate than the Geddes rule with changes is cardiac stroke volume. A performance evaluation showed that the LOP estimation time would be nearly three times faster than that possible with a Critikon Dinamap 845, with an upper limb LOP estimation accuracy in the laboratory which is comparable to that achieved by the Dinamap 1848 in the estimation of SBP. Simple experiments showed that the algorithm is capable of tracking rapid variations in LOP due to changes in limb position or blood pressure. This estimation speed plus the accuracy of the pulse detection technique should make the algorithm effective in the adaptive control of tourniquet pressure in the operating room. Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System 91 CHAPTER 6 DEVELOPMENT AND CLINICAL EVALUATION OF THE ADAPTIVE TOURNIQUET SYSTEM In the previous three chapters, a specif ic aspect of the adaptive tourniquet problem was examined, wh ich led to improvements in cuff design and specialized techniques for artifact rejection, signal detection, and L O P estimation. This chapter oudines the integration of these improvements into a practical adaptive tourniquet system which was used in the orthopaedic surgical treatment of 16 patients at Vancouver General Hospital . The purpose of the c l in ica l study was to first determine whether or not adaptive regulation of tourniquet pressure based on oscil lometric L O P estimation would significantly reduce the l imb-applied pressure, thereby increasing patient safety. The second purpose was to evaluate the overall effectiveness of the system, particularly the signal processing and L O P estimation techniques, in a c l in ical environment. Results of the study could be compared to those obtained with earlier adaptive tourniquet systems [5, 6], and could also be used to suggest practical design improvements for future systems. The chapter begins with a description of the hardware and software which comprised the adaptive tourniquet system. This is fol lowed by a discussion of the approvals and testing required in preparation for the trials and the experimental protocol. Five case reports, each of which illustrates a different and particularly interesting result, are then given. The chapter concludes with a synopsis wh ich summarizes the more significant results obtained from all 16 surgical cases. 6.1 System Descr ipt ion 6.1.1 H a r d w a r e A s with the experimental apparatus, the adaptive tourniquet system was designed around an I B M Personal Computer which had been equipped with the pneumatic and electronic components illustrated in Figures 6.1 and 6.2. The pneumatic system shown in Figure 6.1 is a significant improvement over the system used in the experimental apparatus (Fig. 4.1). The sensor of the tourniquet cuff is connected to two air l ines, one of which supplies air and the other of which acts as a Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System 92 IBM Personal Computer Cuff Sensor (proximal) Cuff Occluder (distal) F igure 6.1: Pneumatics of the adaptive tourniquet system. Data buss 8 clocks Address buss 16, , Port 1 Port2 Port 3 Transducer Data Acquisition amp gam Control Logic -* valve controls IBM Personal Computer enbl. Aspen ATS 1500 Clinician's Interface Unit LOP Display Status LED's Adaptive o Normal Tone Generator Mode switch -k F igure 6.2: Electronics of the adaptive tourniquet system. 93 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System pressure sensing l ine. This arrangement prevents measurement errors from pressure drops caused by air f low through the transducer, and it also permits detection of line kinks and air leaks at the cuff connections. Three valves are used to control the sensor pressure, and this configuration allows zeroing of the transducer without deflating the sensor. This is accomplished by first activating V2 to seal of f the cuff, fol lowed by activation of V3 to vent the transducer to the atmosphere. A n Aspen A T S 1000 tourniquet controller was used as a pressure source to supply the computer pneumatics with a regulated 350 mm H g , and an Aspen A T S 1500 tourniquet controller (Aspen Laboratories, Englewood C O ) was used to pressurize the distal bladder, or occluder, o f the tourniquet cuff. A s shown, the Aspen A T S 1500 received signals from the I B M P C to vary the pressure in the occluder. The electronics system in the I B M P C incorporated the instrumentation ampli f ier, filter, attenuator, and A - D converter as illustrated previously in Figures 4.2 and 5.3, and wh ich appear as the data acquisition block in Figure 6.2. A lso shown are the additional logic elements wh ich provided IO ports, address decoding and control, and two low frequency clocks. A 40 H z clock was used to set the data acquisition sampling rate, and a 400 H z clock was used for deflation valve timing and miscellaneous time-based functions. A s in the experimental apparatus, these components and the pneumatics were assembled on a breadboard and installed in the fifth expansion slot of the computer. The figure shows that the I B M P C communicates pressure data to the Aspen A T S 1500 via a parallel interface. A switch was added to the A T S 1500 which enabled or disabled control from the computer. The required modifications to the A T S 1500 software were supplied by Aspen Labs, Colorado, in the form of two E P R O M S which replaced the standard 1500 firmware. The cl inician's interface unit (CIU) was designed to inform the surgeon about the status of the system. The unit provides a three-digit numeric display of the L O P , L E D ' s to display the system status, and a sound generator with speakers for emitting tones to indicate a change in status. The unit also provides a pushbutton switch for the surgeon or other clinician to enable or disable the adaptive mode of operation. Figure 6.3 is a photograph of the assembled adaptive tourniquet system. The computer appears on the right, situated on a stainless steel cart with wheels. Mounted on an IV pole at the left are the C I U (top), Aspen A T S 1500 (middle) and Aspen A T S 1000 (bottom). The cuff hoses appear coi led at the top of the IV pole, and the electrical connections from the computer to the A T S 1500 and C I U are vis ible between the A T S 1000 and the cart. Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System 94 Figure 63: The adaptive tourniquet system hardware. 6.1.2 Software The software described in this section provides the user interface and control of the external units connected to the computer. The algorithms required for sensor deflation, pulse acquisition, and L O P estimation which were incorporated into this software were previously described in Chapter 5. The adaptive tourniquet program (Fig. 6.4) provides two system modes and several options which can be changed with the user options menu. In constant pressure mode, the computer does not measure L O P , and the Aspen ATS 1500 regulates the tourniquet occluder at the set pressure, which is one of the user-set options. In adaptive pressure mode, the system measures patient L O P , and displays estimates on the C I U . If the closed loop operation option has been selected, the ATS 1500 pressure is set to the current L O P estimate plus the safety offset pressure at the end of each measurement cycle. The safety offset pressure may be set to any integer value with the options menu. If the closed loop 95 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System start self test user options menu Y / x h a n g e X ^ N options transmit set pressure to ATS 1500 constant pressure mode Y ^ c h a n g e ^ ^ N options Y / mode \ N switch adaptive pressure mode step-deflate sensor to obtain system time constant mode switch I LOP estimation algorithm (Fig. 5.16) LOP, transmit L O P e + safety offset to ATS 1500 write LOP to display write data to hard disk Figure 6.4: Adaptive tourniquet system software. 96 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System any integer value with the options menu. If the closed loop operation option is disabled, the system continues to measure and display the L O P in adaptive mode, but the A T S 1500 pressure remains at the set pressure. In adaptive mode, the P C video monitor is used as a real-time display of the tourniquet sensor baseline pressure and oscil lometric pulse signals. In constant pressure mode, the monitor displays text such as the options menu or other screens providing instructions or system diagnostics. The P C keyboard is the primary input device. The system mode may be switched at any time by str iking ' m ' on the keyboard or by pushing the front panel switch on the C I U . From the keyboard, the user can restart the L O P estimation algorithm from the beginning (see F ig . 5.16), deflate the sensor, or quit the program by striking the appropriate key. Unique alarm conditions are declared i f the sensor cannot be inflated due to a l ine k ink or leak, i f the tourniquet inflation time has extended beyond a preset time l imit, or i f the osci l lometric data acquisit ion time becomes greater than approximately 15, 30, or 45 seconds. Except for the cases where L O P estimation time extends beyond 15 or 30 seconds, the system wi l l automatically revert to constant pressure mode in response to these alarms. Alarms are indicated with the status L E D ' s and alarm sound generator of the C I U , and by text on the P C video monitor. The system records information such as mode switches, alarms, patient L O P , and Aspen 1500 pressure in a temporal format in patient data (.PDF) files which are stored on the fixed disk drive. These files provide a header for entering trial-related information such as the hospital, O R suite, surgeon, operation, and patient dimensions. A patient data file from the tenth c l in ical trial has been included as the Appendix. 6.2 Prepara t ions for C l i n i c a l T r i a l s Before any new medical device can be evaluated on patients undergoing treatment at one of the university hospitals, an experimental protocol must be developed, and a proposal submitted and approved by the C l in ica l Screening Committee of the Faculty of Medic ine, Universi ty of Brit ish Co lumbia . The screening committee consists mostly of physicians who examine primari ly the ethical aspects of the proposed experiment or medical device evaluation. Factors considered are the risks and benefits of the experiment to the subject or patient, and whether or not the patient w i l l receive the same quality of health care as he would normally have. 97 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System Following approval on ethical grounds, further approval must be obtained from the Research Committee of the hospital where the experiments are conducted. In this case, the concerns are more administrative, and information which illustrates how the experiment will impact on patient care, clinical staff, and hospital resources must be provided. A n experimental protocol was written for the clinical trials of the adaptive tourniquet system, and the required information was submitted to the screening committees. Approval was obtained from the Faculty of Medicine, Vancouver General Hospital, and the Acute Care Hospital at U B C to conduct the trials under the supervision of Dr. Robert McGraw, professor and head of the Department of Orthopaedics, U B C . Prior to its use in clinical trials, the adaptive tourniquet system design was reviewed by an engineer working independently in the Biomedical Engineering Department of V G H to ensure that any single component failure in the system could be promptly detected and safely corrected. The adaptive tourniquet system was also tested by another independent member of the department to ensure that the device met all pertinent regulations and safety standards. The ATS 1000 and 1500 were inspected and calibrated to bring the units up to factory specifications. Leakage current and other tests were conducted to ensure that the adaptive tourniquet system surpassed the minimum specifications set out in the CSA standard C22.2 number 125, Electromedical Devices [90]. 6.3 Experimental Protocol Suitable patients were selected in accordance with the approved protocol by the surgeons participating in the study. Each trial was conducted under the direction of the surgeon performing the procedure, who had been previously instructed in the operation of the system. In all trials, an experienced biomedical engineer was in constant attendance to collect data, observe, and assist in the operation of the system. The system was located well outside the surgical field (Fig. 6.5) where it was monitored and operated by the biomedical engineer under verbal directions from the surgeon. After anesthetic induction, the limb was wrapped with two or three layers of soft bandage, the cuff was applied, and the limb was draped and prepared for surgery. The set pressure was initialized as 200 - 250 mm Hg for upper limb surgeries and 250 - 300 mm Hg for lower limb surgeries. The safety offset pressure was chosen based on experience and experimental observations such as LOP estimation error and L O P variation with limb positioning. 98 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System Figure 6.5: The adaptive tourniquet system in surgery. (Courtesy Drs. Moran (left), M i l l e r (center), and Beauchamp (right)). Typica l ly , safety offsets of 15 - 25 mm H g were used on the upper l imb and offsets of 30 - 40 mm H g were used on the lower l imb. In some cases, the surgeon permitted the use of a zero safety offset when leakage into the surgical site would not disturb the technique or was in fact desired. A t zero offset, the pressure in the tourniquet occluder equals the L O P estimate, and hence validity of the estimate can be determined by looking for blood in the field. If the procedure was a total knee arthroplasty, a protocol was established in that the closed loop operation option would not be enabled until after the prosthesis had been installed and the bone cement had dried, thereby reducing the risk of blood leakage into the surgical site before the cement had properly cured. A dry surgical field is required to maximize the effectiveness of the bone cement. 6.4 Clinical Trial Results This section presents and interprets the results which were obtained with the system in 16 surgical cases. First, specific cases in which the results illustrate interesting or important points are discussed. 99 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System This is fol lowed by a synopsis summarizing the results of all trials and the conclusions which were obtained from them. 6.4.1 Cases 1 to 3: version 1.2 software The first trials of the adaptive tourniquet system used an early version of the software designated v. 1.2. W i t h this version, the improved sensor deflation and fast L O P estimation algorithms presented in Sections 5.1 and 5.3 respectively had not yet been developed, although a pulse pattern recognition algorithm similar to the one described in Section 5.2 was in operation. In these cases, L O P estimates were obtained with a Geddes rule algorithm, and were not smoothed with a running average before being used to set the distal bladder pressure. This first algorithm reacted conservatively i f the L O P could not be determined f rom the acquired data. For example, i f the tourniquet sensor had not been init ial ly inflated to above L O P and the occlusion pressure not determined, the algorithm assumed that the L O P was equal to the set pressure minus the safety offset, which in closed loop operation would produce a distal bladder pressure equal to the set pressure. This rule would ensure effective hemostasis during closed loop operation i f the algorithm failed to produce an L O P estimate. In the first two cl inical trials, the adaptive tourniquet system was used in total knee arthroplasties performed by Dr. M c G r a w . For the first time, quantitative data substantiating the performance of completely adaptive control of tourniquet pressure on the lower l imb in the cl in ical environment was recorded and is presented in this thesis. Figure 6.6 illustrates the L O P variations observed in the first cl inical trial, in which the patient was an 85 year-old male with osteoarthritis of the left knee. The figure shows the L O P estimates, the pressure in the tourniquet occluder, and the systolic blood pressure (SBP) of the patient. The S B P data points shown at five minute intervals in this and all subsequent examples were obtained from the anesthetist's record. Note that the patient was hypertensive with a mean S B P of 160 mm H g , resulting in relatively high L O P estimates. Despite this, Dr. McGraw noted no blood leakage into the surgical f ield throughout the case. W i th two distinct exceptions, Figure 6.6 shows the L O P estimates remain mostly within +20, -10 m m H g of S B P . The first exception occurs at t = 10 minutes, where, probably due to data corruption from artifact, the algorithm did not locate a pulse with an amplitude of less than 5 5 % of the maximum amplitude in the collected data, and hence overestimated the L O P as the set pressure, 275 mm Hg, minus the safety offset, 30 mm Hg . L O P spikes also appear after t = 84 minutes due to this rule 100 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System operating on data corrupted by repetitive l imb movements wh ich varied from ful l f lexion to full extension. This data shows why this rule was eventually rejected and a running average was applied to smooth the L O P estimates. 350 300 •a 2 5 0 a | 200 <u 3 150 u fc 100 50 0 0 16 32 48 64 80 96 112 Time (minutes) Figure 6.6: Results obtained from first cl inical trial. Surgeon: McGraw . Procedure: Total knee arthroplasty. For approximately 8 < t < 88 minutes, the l imb remained in a ful ly flexed position. Fo l lowing this came the repetitive l imb manipulations noted above to test the stability of the prosthesis, producing noise in the estimates. A t t = 94 minutes the l imb was lowered, and closed loop control was enabled for a total of 16.8 minutes, during which the wound was closed and the l imb bandaged. The mean L O P , wh ich was 205.5 mm Hg for 94 < t < 112 minutes, appears wel l in excess of S B P during this interval. This effect, which has been observed to a greater or lesser extent in all knee arthroplasty trials, could be due to a decrease in snugness caused by a change in cuff position as the l imb was moved from full f lexion to full extension, thereby increasing the L O P . It may be that the cuffs used in these trials gradually decrease in snugness due to elastic stretching of the outer layer for cases which are long or which involve extended periods of l imb movement. Dur ing closed loop control of the occluder pressure, the mean time between estimates was 34 seconds. Since approximately 10 to 12 seconds were required to f i l l the volume of the leg sensor, the Geddes rule algorithm required nominally 22 seconds to acquire the necessary data to estimate L O P , twice the desired acquisition time performance specified at the beginning of Chapter 5. A consequence Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System 101 of this was that in the first three trials, the 15-second data acquisition time alarm was often activated even though signal artifacts were small or nonexistent. These early clinical results were the primary motivation to develop the improved L O P estimation algorithm which was implemented as software version 2.0, and was tested in the laboratory as described in Chapter 5. The method of reporting alarms based on the inter-estimate interval was changed in software version 2.1 to a method based on the number of attempts the pulse acquisition algorithm would make in collecting an uncorrupted set of pulse amplitude data. 6.4.2 Cases 4 to 16: version 2.1 software Figure 6.7 shows the results from the sixth clinical trial, which was a total knee arthroplasty performed this time by Dr. R. D. Beauchamp, Dept. of Orthopaedics, U B C . The results show close tracking of the SBP by the L O P estimates for 10 < t < 60 minutes. At t = 75 minutes, the limb was extended for wound closure, and closed loop operation was enabled for 15.8 minutes. As had been seen in previous cases, L O P remained above SBP during this period, due most probably to a reduction in cuff snugness. 350 300 100 50 0 i i i—r T Conventional pneumatic tourniquet pressure L.O.P. Estimate Occluder Pressure Systolic Blood Pressure — -e— j • 20 40 60 Time (minutes) 80 100 Figure 6.7: Results obtained from sixth clinical trial. Surgeon: Beauchamp. Procedure: Total knee arthroplasty. A comparison of Figures 6.6 and 6.7 shows the effect of three-point smoothing in reducing LOP estimate noise. Mean time between estimates during closed loop control was 19.2 seconds, mean LOP was 164.1 mm Hg, and mean occluder pressure was 204.2 mm Hg, a 96 mm Hg reduction from the 102 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System conventional lower l imb tourniquet pressure of 300 mm Hg . N o leakage into the surgical wound was noted in this case. Figure 6.8 shows the results from the first upper l imb trial, an internal reduction and f ixation of the left ulna, also performed by Dr. Beauchamp. The patient had a smal l , tubular upper l imb with a circumference of only 23.5 cm, and consequently the cuff was readily applied with a high degree of snugness. This is probably why the L O P of the patient was 10 - 20 mm H g less than S B P for the first 25 minutes of the case. 300 • i , i I | I I I I | ! I I 1 | I I I I j | | | | | | | | ! Conventional pneumatic tourniquet pressure 250 '- : 3 200 -; L.O.P. Estimate 5 0 -_ Occluder Pressure Systolic Blood Pressure a 0 I i i i i 1 ' 1 ' ' 1 ' 1 1 1 1 ' ' 1 • i i i i i i . i . i 0 10 20 30 40 50 60 Time (minutes) F i gu re 6.8: Results obtained from seventh cl inical trial. Surgeon: Beauchamp. Procedure: Internal reduction and fixation of ulna. The L O P estimation algorithm was started at t = 4 minutes, and closed loop control was enabled at t = 12 minutes. Adaptive control o f the distal bladder pressure was in effect until the end of the case 50.2 minutes later, the longest period over which closed loop control has been permitted in any trial. A s discussed in Section 5.3.2, smoothing is not applied to sudden increases in L O P over 15 mm Hg, and spikes appear in the L O P data at t = 22 minutes and 41 < t < 47 minutes. The origin of the first spike is unknown, however, the later spikes were caused by Dr. Moran, who, when assisting Dr. Beauchamp, was leaning on the tourniquet cuff sensor while holding up the forearm. Eventually, the system automatically switched from adaptive to constant pressure mode in response to these artifacts (not illustrated), and was restarted manually shortly thereafter. For this case, the mean inter-estimate interval was 7.2 seconds during closed loop control of the occluder pressure. Mean L O P was estimated at 120.4 mm Hg , and the mean occluder pressure was 103 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System 150.0 mm Hg, 100 mm Hg less than the 250 mm Hg conventionally used in upper limb surgery. No leakage into the surgical site was noted in this case. Results of the tenth clinical trial, a biopsy of the left talus performed by Dr. R. Claridge, are shown in Figure 6.9. These lower limb LOP's are highly unusual in that some of them are up to 20 mm Hg less than the SBP of the patient. These results are difficult to explain, although the patient did have a tubular limb geometry with a small circumference of only 44 cm. The cuff width-to-limb circumference ratio was therefore larger than for the average patient, which could partially explain the low LOP values observed. 350 I i i i i | i i i i | i i i—i | i i i i | i i—i i | i i i i | i I i i Conventional pneumatic tourniquet pressure 300 : : L.O.P. Estimate '-50 - Occluder Pressure -Systolic Blood Pressure Q -0 r i . . . i . • • • i i • • • I i i * > I t i > • I i • • • I i i i i " 0 5 10 15 20 25 30 35 Time (minutes) Figure 6.9: Results obtained from tenth clinical trial. Surgeon: Claridge. Procedure: Talus biopsy. During closed loop control over 6 < t < 36 minutes, mean inter-estimate time was 15.6 seconds, mean LOP was 115.9 mm Hg, and mean occluder pressure was only 155.9 mm Hg. Despite this low tourniquet pressure, blood was never observed in the surgical site. Figure 6.10 is the final example illustrating the results obtained during a metacarpal arthrodesis (MCA) and tendon repair performed by Dr. P. Gropper, Dept. of Orthopaedics UBC, on the right hand of a 47 year-old female in the 12th clinical trial. At t = 25 minutes, the set pressure was reduced from 250 to 200 mm Hg, and at t = 43 minutes, closed loop operation was enabled with a safety offset of 50 mm Hg. Under the instructions of Dr. Gropper, the safety offset was reduced to 40, 20, 10, and finally to zero mm Hg at t = 67 minutes. Even at zero offset, no blood was observed in the surgical field, and Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System not unti l the offset was further reduced to -15 mm Hg did the wounds become engorged wi th blood. A t this point, the tourniquet occluder pressure had fallen below S B P by 10 mm Hg . 22 32 42 52 62 72 Time (minutes) F i g u r e 6.10: Results obtained from 12th cl inical trial. Surgeon: Gropper. Procedure: Metacarpal arthrodesis. Th is was another unanticipated result which was not repeated when a similar experiment was conducted in a wrist arthrodesis, also performed by Dr . Gropper. In this trial, blood leakage was observed at the end of the case when the safety offset pressure was reduced from 20 to 0 mm Hg. The result observed in the previous M C A , however, could be attributed to exsanguination of the limb prior to surgery, which has been demonstrated to depress the L O P below the value observed when the venous reservoir remains f i l led with blood [23]. 6.4.3 Synopsis Table 6.1 summarizes the results obtained during closed loop control of the distal bladder pressure from 11 of the cl in ical trials. The first three trials were excluded from this summary since they were conducted with version 1.2 of the software. A lso excluded were the results from trials 5 and 15 due to problems which prevented the use of closed loop operation. These problems are discussed in detail fo l lowing the summary. The table segments the results into upper and lower l imb surgical cases. The total closed loop operation time and number of L O P estimates were calculated by summing these quantities over all 105 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System relevant trials. The means shown were obtained by first calculating the average of the parameter for a given trial, fo l lowed by summing the averages and dividing by the number of trials. Summary of Upper Limb Trials - 4 surgeries Total closed loop control time (min) Total number of LOP measurements Mean LOP estimation time (sec) Mean LOP (mm Hg) Mean occluder pressure (mm Hg) Conventional tourniquet pressure (mm Hg) 116.0 799 9.6 135.6 162.0 250 Summary of Lower Limb Trials - 7 surgeries Total closed loop control time (min) Total number of LOP measurements Mean LOP estimation time (sec) Mean LOP (mm Hg) Mean occluder pressure (mm Hg) Conventional tourniquet pressure (mm Hg) 178.9 737 18.6 148.9 186.7 300 Tab le 6.1: Summary of observations during closed loop control of distal bladder pressure in 11 cl inical trials with software version 2.1. The first point to note is that the use of this adaptive control technique reduced mean tourniquet pressure below conventional standards by 113 mm H g on the lower l imb and 88 mm H g on the upper l imb. A l though this is not shown in the table, the difference between the commonly used tourniquet pressure and the average l imb-applied pressure ranged from 70 to 144 mm Hg across all eleven cases. These reductions are significant, so that the risk of a tourniquet-related injury was probably decreased through use of the system. The lower pressures encountered were due mostly to improved cuff design, and results o f this nature were in fact predicted in Chapter 3. The second point to note is that the mean estimation time for lower limb cases is approximately twice that observed in the upper l imb surgeries. This is simply because, as was discussed in the acquisit ion algorithm design presented in Section 5.2, deflation of the lower l imb sensor requires so much time that the first pulse after the deflation pressure step is obscured, and the algorithm must wait for the second pulse. Consequently, L O P estimation requires twice as many heartbeats on the lower l imb as on the upper l imb. A s evidenced by Figures 5.18, 6.7 and 6.9, these longer estimation times did not apparently degrade L O P tracking performance. However, the lower limb L O P estimation time 106 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System can be reduced such that it equals the upper l imb estimation time by making a simple improvement to the pneumatic system. This improvement is discussed in the fol lowing chapter. Another point to note here is that the L O P estimation time for the upper l imb is significantly longer than that predicted in Section 5.3.3.1. Th is is because of two reasons. First, the mean L O P estimation time was computed from data which included all interruptions of adaptive mode control by either manual or automatic switching of the system from adaptive to constant pressure mode. These mode switches arose from excessive artifacts or were intentional so that options such as the safety offset pressure could be varied. Second, these observations include the time required to inflate the sensor in addition to the pulse acquisition time. Sensor inflation was not considered in the estimation time calculations o f Section 5.3.3.1. M c E w e n d id not report a significant amount of c l in ical trial data [5]. However, as his implementation used a Cr i t ikon Dinamap 845 to estimate L O P once per minute, the implementation evaluated here improved on this aspect by at least a factor of three in lower l imb surgery and by at least a factor of six in upper l imb surgery. It is diff icult to compare the effectiveness of system evaluated here to the one tested by Bussani, since, in the latter case, closed loop control o f the tourniquet pressure was never used, and hence the effectiveness of Bussani 's system in maintaining hemostasis during an operation was never determined [6]. However, Bussani reported a mean L O P estimation time of 12.5 seconds during surgery for his implementation, which is slightly worse than that achieved by this system on the upper l imb, but which is 50% better than that achieved in the lower l imb surgeries. As noted above, lower l imb L O P estimation time for the existing system can be halved with a simple pneumatic modification. Bussan i 's system used a tourniquet cuff which was much narrower than the cuff used in these trials, and consequently he observed L O P estimates which were somewhat higher. A s mentioned previously, the adaptive tourniquet system did not perform perfectly in all trials. In the 15th trial, adaptive mode could not be used because the oscillometric signal observed from the patient's upper l imb had an amplitude which saturated the A - D converter. Even at maximum signal attenuation, the pulses were clipped and rejected as artifacts simply because the patient's pulse pressure was 70 mm Hg , an extreme value which was totally unanticipated in the design o f the analog attenuator. B lood leakage into the surgical wound occurred in trials 5, 8 and 16, for reasons which remain a mystery considering the excellent results obtained in the other 13 trials. Fortunately, this leakage did 107 Chapter 6: Development and Clinical Evaluation of the Adaptive Tourniquet System not affect the outcome of the procedures for the patients involved. In trial 5, leakage occurred inexplicably while the occluder was still at a pressure of 280 mm H g , and hence closed loop operation was never enabled. In the eighth trial, some leakage occurred while operating in closed loop as wound closure was approaching completion. In this case, leakage may have been due to an inappropriate choice of only 20 mm H g for the safety offset pressure. The sixteenth trial was an upper l imb case involv ing a geriatric patient, and leakage here could possibly be due to calcif icat ion of the brachial artery located underneath the tourniquet occluder. Despite these problems, the majority of the results showed that the integration of the individual improvements i n cuff design, oscil lometric signal processing, and L O P estimation produced a safe and effective system which can now be evaluated on a continous basis in the operating room. This was not achieved by any previous implementation. Under the provisions of the hospital and screening committee approvals, further cl inical evaluation of this system to determine the benefits realized through the use of an adaptive tourniquet w i l l continue from now until at least July 1992. T o a more l imited extent, the cl inical trials have also illustrated that adaptive control o f the tourniquet pressure should improve patient safety through a significant reduction in the l imb applied pressure. However, to what degree the risk of compression injury is minimized or the rate of postoperative complications is reduced remains unknown. Because l i tde, i f any, L O P monitoring instrumentation can be used during surgery, it is difficult to determine the accuracy of the system in estimating L O P in the cl in ical environment. However, in addition to the medical information which can be gained from evaluations l ike these, cl in ical trials can also be used to identify problems with the implementation which would lead to practical design improvements. In the next chapter, practical problems which were found to be common to a large number of cases are identified and their solutions proposed. Chapter 7: Conclusions and Suggestions for Further Research 108 CHAPTER 7 CONCLUSIONS AND SUGGESTIONS FOR FURTHER RESEARCH 7.1 Conclusions Many patient injuries which result from the use of pneumatic tourniquets are known to be related to cuff inflation pressure. Previous research showed that the use of wider cuffs would permit lower inf lat ion pressures to achieve occlusion, thereby reducing the risk o f postoperative complications such as sensory and motor loss from underlying neural damage. Paral lel investigations concerned the development of adaptive tourniquet pressure controllers which would regulate the cuff pressure near the min imum required to occlude the l imb vessels for a given duration. Practical problems plagued early adaptive tourniquet implementations which precluded their extensive cl in ical use. The purpose of this research was to investigate problems in adaptive tourniquet control which had previously been identif ied, but the extent and character of which were not wel l understood analytically. These investigations, which led to contributions in cuff design, oscil lometric signal processing, and l imb occlusion pressure estimation, culminated in the development o f a practical system which could be extensively used in the treatment of patients undergoing orthopaedic surgery. The first problem investigated was how to ensure the pressure inside the cuff bladder d id in fact reach the l imb surface. Quantitative criteria were used in the design of a wide, non-rigid cuff which achieved an improved fit to the l imb, thereby reducing the sensitivity o f the pressure transmission to snugness differences caused by geometric variations from patient to patient. Based on preliminary results, a comparison of the new design and a conventional dual-bladder tourniquet showed that the new cuf f would be capable of providing improved performance and reliability not only as an occlusion effector, but as an oscil lometric occlusion sensor as wel l . The next problem investigated was how to economically reduce the disruptive effects o f surgical manipulations on the signals used for L O P estimation. A detailed study of oscil lometric pulse and noise signals eventually led to the conclusion that a sophisticated digital signal processing approach was not warranted in solving the noise problem, a highly unconventional result considering the current popularity o f digital techniques. A simple analog filter was proposed for suppressing the manipulation 109 Chapter 7: Conclusions and Suggestions for Further Research noise which was analytically shown to perform nearly as wel l as all patient-specific optimum digital filters on average. However, observation of the filtered signals during a procedure showed that many other signal artifacts existed besides l imb movement. Observations l ike these led to the development of a heuristic scheme for detecting the filtered osci l lometr ic pulses under artifactual conditions. The scheme uses the amplitude, slope, and timing features o f previously acquired pulses to calculate the detection thresholds for the next pulse in the acquisit ion cycle. A n evaluation of this approach on corrupted signal data acquired during an operation showed a correct detection performance of 83%, with false positive and false negative error performances of 2 3 % and 17% respectively. Ear ly c l in ical results wi th the adaptive tourniquet system showed long L O P estimation times. This inspired an analytical investigation of the biomechanics o f oscillometry v ia computer simulation, wh ich resulted in a new, faster method for estimating L O P . A primary advantage of the method is that, unl ike al l previous approaches, the cuff pressure required to estimate the L O P is subocclusive, thereby virtually el iminating any risk of patient injury due to occlusion pressure sensing. Satisfactory results were obtained when the method was tested both by computer simulation and in laboratory experiments. Sixteen cl in ical trials were conducted with an adaptive tourniquet system which integrated these individual improvements, and to a large extent the trials were successful. For the first time, quantitative data were obtained during lower l imb surgery illustrating the performance of an adaptive tourniquet system during closed loop control of the cuff pressure. In one trial, the system failed to measure L O P due to an unforeseen limitation in the electronic hardware. In three out of the sixteen trials, blood leaked past the cuff into the surgical wound for reasons which remain either partially or completely unknown. However, the other 13 trials indicated that the contributions of this research resulted in a medical device which can now be evaluated on a routine basis in the operating room. This was not accomplished with any previous system. The results from 11 of the trials also showed that, on average, use of the system in closed loop operation significantly reduced the average limb-applied pressure by 35%, or from 250 to 162 mm Hg , for upper l imb surgeries, and by 38%, or from 300 to 187 mm H g , for lower l imb surgeries. Use of these lower pressures should reduce the risk of tourniquet-related injuries, however, the impact of adaptive regulation o f tourniquet pressure on the actual incidence of these injuries remains a subject for further cl inical study. 110 Chapter 7: Conclusions and Suggestions for Further Research 7.2 Practical Improvements and Suggestions for Further Research Dur ing the course of this work, various problems in addition to the ones investigated in detail were identif ied, some of which are technical and require practical solutions, and some of which are more fundamental and require substantial scientific research. Most of the practical problems were identif ied during the cl inical trials, and these are discussed first, fol lowed by a presentation of recommendations for further research. 7.2.1 Practical improvements 7.2.7.7 Cuff design The cuffs designed for this system were adequate for this initial phase of research, and must be improved before further cl in ical testing can proceed. A n improved cuff bracket must be designed which w i l l simplify application of the cuff in the cl inical environment, especially for lower l imb surgery. Alternative fiber mesh /PVC composite materials for cuff construction should be investigated to develop designs which do not stretch with pressure cycl ing and l imb manipulation. A more detailed anthropometric data base specifically for l imb dimensions should be compiled which lists not only lengths, circumferences and tapers, but also the correlations between these parameters so that a better understanding of the range of geometric variations can be obtained. This knowledge would permit the development of cuff application methods which would simplify or improve on the cuff-and-clamp approach presented here. Fortunately, the patients and staff o f a major hospital form a large, highly varied population which would permit such anthropometric data to be collected rapidly. 7.2.1.2 Pneumatic design A s was discussed in Chapter 6, the data acquisition time for L O P estimation on the lower l imb was twice that for the upper l imb simply because the large volume of the leg sensor could not be deflated quickly enough between heartbeats. This may be remedied by paralleling the deflation valve V 3 in Figure 6.1 with a second valve that has a larger deflation orifice. If the valve-on time of V 3 to achieve a desired pressure step exceeds, say, 250 msec, the valve with the larger orifice would then be chosen to deflate the cuff. In order to calculate the valve-on times in this case, the step-deflation algorithm would require two resistance-compliance time constants, one for each deflation valve. I l l Chapter 7: Conclusions and Suggestions for Further Research 7.2.1.3. Electronic design Early in the development of the adaptive tourniquet system, it was determined that the dynamic range of the oscillometric signal was significandy greater than that of an 8-bit A-D converter, and hence a dynamic range compression scheme based partially in hardware and partially in software was implemented to overcome this limitation. However, the appropriate selection of the A-D converter resolution would eliminate this bothersome complexity, and would result in improved performance at reduced cost since higher resolution, low speed converters are readily available. Simple calculations reveal that a 12-bit A-D converter would resolve even the smallest observed oscillometric signal into 100 parts. A second improvement in the hardware design is of course that the elements should be assembled into a single unit for improved portability. Except for the few additional components required to process the oscillometric signal and to provide a user interface, a conventional tourniquet controller like the Aspen ATS 1500 provides all the hardware necessary to implement such a device. The most significant difference between an adaptive and a non-adaptive tourniquet is that the latter requires extensive software additions to perform oscillometric data acquisition. 7.2.2 Topics for further research 7.2.2.1 Engineering research Although the improvements in oscillometric signal processing and LOP estimation presented in this thesis were developed with a surgical application in mind, all of these techniques may be applied to the much more general problem of noninvasive blood pressure estimation. The artifact suppression and signal pattern recognition methods developed here may be applied to improve the performance of automated oscillometric instruments in situations where limb movement would be a problem, for example, in ambulatory patient monitoring, physiological stress testing, and aerospace medicine. Although the new method of LOP estimation could not be used to continuously monitor the blood pressure due to the effects of venous congestion, it could be useful in situations where noninvasive observations of systolic pressure dynamics in response to brief stimuli are required over a period of one or two minutes. An example of this would be the Valsalva maneuver experiment illustrated in Figure 5.19. Further research can be conducted to adapt these filtering, detection, and LOP estimation techniques to more general problems in oscillometric blood pressure estimation. Having completed 112 Chapter 7: Conclusions and Suggestions for Further Research this, the degree to which these techniques improve performance over that obtained by conventional methods can be evaluated. Al though the problem of processing the oscil lometric signal to suppress artifact was analytically investigated, the pattern recognition problem was not. Limitat ions were encountered with the heuristic detection algorithm such as a false positive error which became greater due to high levels of artifact. In more recent cl inical trials conducted at this time of writ ing, another problem has appeared in that some patients do not produce a simple, triangular waveform such as the one illustrated in Figure 5.6, and as a result the algorithm rejects the first pulse in the acquisition cycle since its fal l ing edge does not meet the slope and amplitude criteria defined for validity. Consequendy, to achieve improved performance, further research should be devoted to this problem perhaps along more traditional lines. For example, an approach based on discriminant functions could be compared to heuristic, ad hoc schemes. The development of more sophisticated techniques would require a detailed analysis of the stochastics o f both signal and noise features, implying a need for a larger study involv ing a wider variety of subjects. The heuristic scheme developed here could be considered a crude form of predictor, and hence autoregressive predictors could also be evaluated as estimators of the osci l lometric signal amplitude. A s discussed above, results of this research could be applied to both the l imb occlusion pressure and blood pressure estimation problems. The need for a wider variety of subjects in optimizing the oscil lometric pattern recognition rules is also necessary to more ful ly validate the results presented for the cuff surface pressure distributions and the new L O P estimation technique. In this research, the scope of the project and the time constraints placed upon it l imited extensive laboratory evaluation. Investigations which expand upon the work presented here should incorporate experiments with larger populations within their scope. 7.2.2.2 Physiological research Fundamental questions remain concerning the effects of tourniquet pressure, inflation time, and surface pressure distribution on underlying neural damage, specifically the degree of invagination at the nodes of Ranvier or the number of fibers with invaginated nodes. No quantitative information exists relating tourniquet parameters to neural damage in humans, and thus far only a qualitative understanding has been developed based mostly on animal studies and sporadic surgical case reports. A s was mentioned previously, the introduction of various technical improvements such as better cuff designs and adaptive control of the tourniquet pressure intuitively seem capable of reducing both the 113 Chapter 7: Conclusions and Suggestions for Further Research degree and extent of tourniquet injuries, although there is little i f any quantitative evidence to substantiate this c la im. B iomedica l engineers can participate in these fundamental pathological investigations by developing the specialized instrumentation required to quantify tourniquet injury. This is an exceptionally diff icult problem due to the extreme sensitivities which would be required. However, without this basic research the benefits of technical improvements in tourniquet design w i l l never be truly substantiated, which in itself w i l l l imit the wide cl in ical acceptance of these technical developments in improving the quality o f patient care. 114 REFERENCES [I] R. W . M c G r a w and J . A . M c E w e n , "The tourniquet," in Unsatisfactory Results in Hand Surgery, Robert M . McFar lane (editor). Edinburgh: Churchi l l L iv ingstone, 1987, pp. 5 - 1 3 . [2] Raoul Tubiana, "The use of the tourniquet," in The Hand, Vol. 2, Raoul Tubiana (editor). Phi ladelphia: W . B . Saunders, 1985, pp. 31 - 37. [3] J . A . M c E w e n , "Complicat ions of and improvements in pneumatic tourniquets used in surgery," Med. Instrumen., vo l . 15, no. 4, pp. 253 - 257, 1981. [4] J . Ochoa, T. J . Fowler, and R. W . Gil l iatt, "Anatomical changes in peripheral nerves compressed by a pneumatic tourniquet," J. Anatomy, vo l . 113, no. 3, pp. 433 - 455, 1972. [5] J . A . M c E w e n and R. W . M c G r a w , " A n adaptive tourniquet for improved safety in surgery," IEEE Trans. Biomed. Eng., vo l . B M E - 2 9 , no. 2, pp. 122 - 128, 1982. [6] C . R. Bussani and J . A . M c E w e n , "Improved tracking of l imb occlusion pressure for surgical tourniquets," IEEE Trans. Biomed. Eng., V o l . B M E - 3 5 , no. 4, pp. 221 - 228, 1988. [7] C . R. Bussani , "Improved noninvasive determination of blood pressure by oscillometry in the presence of motion artifacts," M . A . S c . Thesis, University of Br i t ish Co lumbia , Vancouver, Canada, July 1986. [8] G . McConne l l , "The use of impedance plethysmography to determine the onset of blood flow beneath a tourniquet cuff," M . A . S c . Thesis, University of Br i t ish Columbia, Vancouver, Canada, Ap r i l 1988. [9] G . McConne l l and J . A . M c E w e n , "Determination of the onset of blood f low beneath a tourniquet cuff by impedance plethysmography", Proc. 13th CMBES Conf, June 1987, pp. 53, 54. [10] J . A . M c E w e n , "Occlusive cuff," U.S. Patent 4 ,770, 175, Sept. 13, 1988. [ I I ] J . A . M c E w e n , P. T. Gropper, and R. W . M c G r a w , "New finger cuffs for use with digital tourniquets,",/ . Hand Surg., v o l . l 3 A , no. 6, pp. 888 - 892, 1988. [12] J . A . M c E w e n and G . F. Auchinleck, "Advances in surgical tourniquets," AORN J., V o l . 36, no. 5, pp. 8 8 9 - 896, 1982. [13] J . A . M c E w e n , "Pneumatic tourniquet," U . S. Patent 4, 469,099, Sept. 4, 1984. [14] H. Cushing, "Pneumatic tourniquets: Wi th especial reference to their use in craniotomies," Medical News, no. 84, pp. 577 - 580, 1904. [15] E. M . B rown and T. J . McGr i f f , "Intravenous regional anaesthesia (Bier 's block): Review of 20 years experience," Can. J. Anaesth., vol. 36, no. 3, pp. 307 - 310, 1989. [16] A . H . Crenshaw, "Surgical techniques," in Campbell's Operative Orhopaedics, A . H. Crenshaw (editor). St. Louis: C . V . Mosby, 1987, pp. 3, 4. 115 [17] M . W . Chapman, "Use of the tourniquet," in Operative Orthopaedics, Michae l W . Chapman (editor). Phi ladelphia: J . B . Lippincott, 1988, pp. 76 ,77 . [18] L . L . Johnson, Arthroscopic Surgery: Principles and Practice, Vol. 1. St. Lou is : C . V . Mosby , 1986, p. 116. [19] B. Hara, Complications in Foot Surgery: Prevention and Management. Balt imore: Wi l l i ams and Wi lk ins , 1976, pp. 84, 85. [20] U . T. Larsen and P. Hommelgaard, "Pneumatic tourniquet paralysis fol lowing intravenous regional analgesia," Anaesthesia, vo l . 42, pp. 526 - 528, 1987. [21] A . J . Hodgson, " A parametric study of the axisymmetric loading of an anisotropically elastic l imb model by a tourniquet," M . A . S c . Thesis, University of Bri t ish Columbia, Vancouver, Canada, August 1986. [22] S. Callaghan, " A biomechanical analysis of l imb compression induced by a pneumatic surgical tourniquet," M . A . S c . Thesis, University of Br i t ish Columbia, Vancouver, Canada, Sept. 1989. [23] M . J . Breault, " A biomechanical investigation of blood f low occlusion achieved with the use of surgical pneumatic tourniquets," M . A . S c . Thesis, University of Br i t ish Columbia, Vancouver, Canada, Oct. 1988. [24] M . J . Breault and P. Romi l ly , "Experimental measurement and prediction of blood f low occlusion achieved using pneumatic tourniquets," Proc. 11th Annual International Conf. IEEEIEMBS Society, Nov. 1989, pp. 1604 - 1605. [25] M . R. Moore, S. R. Garf in, A . R. Hargens, "W ide tourniquets eliminate blood f low at low inflation pressures," J. Hand Surg., vo l . 12A, no. 6, pp. 1006 - 1011, 1987. [26] K . Yamakoshi , P. Rol fe, and C . Murphy, "Current developments in non-invasive measurement of arterial blood pressure," / . Biomed. Eng., vo l . 11, no. 2, pp. 130 - 137, 1989. [27] L . A . Geddes, "B lood pressure: Noninvasive measurement," in Cardiovascular Devices and Their Applications. New York: Wi ley and Sons, 1984, pp. 63 - 99. [28] H . P. App le , "Automatic noninvasive blood pressure monitors," in Essential Noninvasive Monitoring in Anesthesia, J . S. Gravenstien, R. S. Newbower, A . K. Ream, and N . T. Smith (editors). New York : Grune and Stratton, 1980, pp. 16 - 23. [29] R. A . Peura, "B lood pressure and sound," in Medical Instrumentation: Application and Design, J . G . Webster (editor). Boston: Houghton M i f f l i n , 1978, pp. 372 - 384. [30] D. A . Paulus, "Noninvasive blood pressure measurement," Med. Instrumen., vo l . 15, no. 2, pp. 91 - 9 4 , 1981. [31] L. A . Geddes and A . G . Moore, "The efficient detection of Korotkov sounds," Med. Biol. Eng., vol. 6, no. 6, pp. 603 - 609, 1968. [32] J . Fraden, "Appl icat ion of piezo/pyroelectric films in medical transducers," / . Clin. Eng., vo l . 13, no. 3, pp. 133 - 138, 1988. 116 [33] J . E . West and I. J . Busch-Vishniac, " F o i l electret transducer for blood pressure monitoring," /. Acoustical Soc. Am., vo l . 74, no. 3, pp. 680 - 686, 1983. [34] S. G . Blank, J . E . West, F. B. Mul ler , R. J . Cody, G . A . Harshf ield, M . S. Pecker, J . H . Laragh, and T. J . Picker ing, "Wideband external pulse recording during cuff deflation: a new technique for the evaluation of the arterial pulse and measurement of arterial blood pressure," Circulation, vo l . 77, no. 6, pp. 1297 - 1305, 1988. [35] J . G . Webster, "Measurement of f low and volume of blood," i n Medical Instrumentation: Application and Design, J . G . Webster (editor). Boston: Houghton M i f f l i n , 1978, pp. 427 -429. [36] D. E . Bahr and K. R. Clark, "Continuous arterial tonometry," in Essential Noninvasive Monitoring in Anesthesia, J . S. Gravenstien, R. S. Newbower, A . K. Ream, and N . T. Smith (editors). New York : Grune and Stratton, 1980, pp. 24 - 29. [37] K . Yamakoshi , H . Shimazu, and T. Togawa, "Indirect measurement of instantaneous arterial blood pressure in the human finger by the vascular unloading technique," IEEE Trans. Biomed. Eng., vo l . B M E - 2 7 , no. 3, pp. 150 - 155, 1980. [38] M . Ramsey III, "Non-invasive automatic determination of mean arterial pressure," Med. Biol. Eng. Comput., vo l . 17, no. 1, pp. 11 - 18, 1978. [39] J . Looney Jr., "B lood pressure by oscil lometry," Medical Electronics, A p r i l , pp. 57 - 63, 1978. [40] W . Silas et al . , "C l in ica l evaluation of the Dinamap 845", Br. Heart J., vo l . 43, no. 2, pp. 202 -205, 1980. [41] M . Yelderman and A . K. Ream, "Indirect measurement of mean blood pressure in the anesthetized patient," Anesthesiology, vo l . 50, no. 3, pp. 253 - 256, 1979. [42] L. A . Geddes, M . Vo lez , C . Combs, D. Reiner, C . F. Babbs, "Characterization of the oscil lometric method for measuring indirect blood pressure," Annals of Biomed. Eng., vol. 10, pp. 271 - 280, 1982. [43] J . A . M c E w e n , "Adaptive pneumatic tourniquet," U.S. Patent 4, 479, 494, Oct. 30, 1984. [44] L. A . Geddes and S. J . Whistler, "The error in indirect blood pressure measurement with the incorrect size of cuff," Am. Heart J., vol . 96, pp. 4 - 8, 1978. [45] R. W . Hornbeck, Numerical Methods. Englewood Cl i f fs N J : Prent ice-Hal l , 1975, pp. 122 -125. [46] M . Schwartz and L. Shaw, Signal Processing: Discrete Spectral Analysis, Detection, and Estimation. New York : M c G r a w - H i l l , pp. 13 - 19. [47] N . Diffrient, A . R . T i l ley, and D. Harmon, Humanscale 71819. Cambridge: M I T Press, 1981. [48] L . W . Sahley, Dimensions of the Human Figure, Male and Female. Cleveland: Cleveland Designers and Consultants Inc., 1957. [49] S. T. Pheasant, Bodyspace: Anthropometry, Ergonomics, and Design. London: Taylor and Francis, 1986. 1 1 7 [50] S. Hoppenfeld and P. deBoer, Surgical Exposures in Orthopaedics. Philadelphia: Lippincott, 1984, pp. 55, 63 ,78 , 79, 83, 88, 98, 391, 396,401, 417, 420,427, 437. [51] A . H . Crenshaw, pp. 90, 1158, 1545, 1545 - 1549. [52] J . A . M c E w e n , "B iomedica l pressure transducer," U . S. Patent Appl icat ion 07 - 033, 770, f i led Ap r i l 3, 1987. [53] R. A . Puera, pp. 340 - 350. [54] W . H . Olson, "Bas ic concepts of instrumentation," in Medical Instrumentation: Application and Design, J . G . Webster (editor). Boston: Houghton Mi f f l i n , 1978, pp. 31 - 35. [55] S. M . K a y and S. L . Marple, Jr., "Spectrum analysis - a modern perspective," Proc. IEEE, vo l . 69, no. 11, pp. 1380- 1419, 1981. [56] F. J . Harr is, " O n the use of windows for harmonic analysis with the discrete Fourier transform," Proc. IEEE, vo l . 66, no. 1, pp. 51 - 83, 1978. [57] A . V . Oppenhiem and R. W . Schafer, Digital Signal Processing. Englewood Cl i f fs N J : Prent ice-Hal l , 1975, pp. 532 - 571. [58] Schwartz and Shaw, pp. 148 - 215. [59] W . H . Press et al . , Numerical Recipes in C: the Art of Scientific Computing. Cambridge: Cambridge University Press, 1988, pp. 407 - 418. [60] J . Makouh l , "L inear prediction: a tutorial review," Proc. IEEE, vo l . 63, pp. 561 - 580, 1975. [61] R. W . Hornbeck, pp. 91 - 98. [62] Schwartz and Shaw, pp. 343 - 349. [63] H . Stark and F. B. Tuteur, Modern Electrical Communications Theory and Systems. Englewood Cl i f fs N J : Prentice-Hall, 1979, pp. 465 - 488. [64] H . Taub and D. L. Schi l l ing, Principles of Communication Systems. New York : M c G r a w - H i l l , 1986, pp. 446 - 452. [65] S. Hayk in , Communication Systems. New York: Wi ley and Sons, 1978, pp. 535 - 543. [66] M . Schwartz, Information Transmission, Modulation, and Noise. New York : M c G r a w - H i l l , 1970, pp. 4 1 0 - 4 2 1 . [67] Dinamap Adult/Pediatric Vi ta l Signs Monitor Service Manual , Cr i t ikon Inc., Tamapa, F L , July 1983. [68] E . A . Gui l lemin, Synthesis of Passive Networks: Theory and Methods. New York: Wi ley and Sons, 1957. [69] Oppenhiem and Schafer, pp. 206 - 211. [70] D o n Lancaster, Active Filter Cookbook. Indianapolis: Howard W . Sams, 1975, pp. 70 - 85. 118 [71] J . L . Seagard, Z . J . Bosnjak, F. A . Hopp, Jr., K. J . Kotr ly , T. J . Ebert, and J . P. Kampine, "Cardiovascular effects o f general anesthesia," in Effects of Anesthesia, B . G . Cov ino , H . A . Fozzard, K . Rehder, and G . Strichartz (editors). Bethesda M D : Amer ican Physiological Society, 1985, pp. 149 - 177. [72] J . G . Reves and S. Gelman, "Cardiovascular effects of intravenous anesthetic drugs," in Effects of Anesthesia, B. G . Cov ino , H . A . Fozzard, K. Rehder, and G . Strichartz (editors). Bethesda M D : Amer ican Physiological Society, 1985, pp. 179 - 193. [73] M . D. Vickers, H . Schmieden and F. D. Wood-Smi th , Drugs in Anesthetic Practice. London: Butterworth, 1984, pp. 320 - 327. [74] R. D. M i l l e r and N . T y Smith, "Neuromuscular blocking agents," in Drug Interactions in Anesthesia, N . Ty Smith and A . N . Corbascio (editors). Philadelphia: Lea and Febiger, 1986, pp. 363 - 390. [75] W . D. Perry and D. H . Heibn, "Linear, low noise inflation system for blood pressure monitor," U . S. Patent 4, 832,039, May 23, 1989. [76] J . Gotman and P. Gloor, "Automatic recognition and quantification of interictal epileptic activity in the human scalp E E G , " Electroenceph. and Clin. Neurophys., vol . 41, pp. 513 -529, 1976. [77] J . R. Glover Jr., P. Y . Ktonas, N . Raghavan, J . M . Urunela, S. S. Velamur i , and E. L . Re i l l y , " A multichannel signal processor for the detection of epileptogenic sharp transients i n the E E G , " IEEE Trans. Biomed. Eng., vol . B M E - 3 3 , no. 12, pp. 1121 - 1128, 1986. [78] P. S. Hamil ton and W . J . Tompkins, "Quantitative investigation of Q R S detection rules using the M I T / B I H arrhythmia data base," IEEE Trans. Biomed. Eng., vol. B M E - 3 3 , no. 12, pp. 1157 - 1165, 1986. [79] J . R. Cox , F. M . Nol le , and R. M . Arthur, "Digi tal analysis of the electroencephalogram, the blood pressure wave, and the electrocardiogram," Proc. IEEE, vol . 60, no. 10, pp. 1137 -1164, 1972. [80] C . L . Feldman, "Trends in computer E C G monitoring," in Trends in Computer-Processed Electrocardiograms, J . H. van Bemmel and J . L. Wi l lems (editors). Amsterdam: North-Hol land, 1977, pp. 3 - 10. [81] F. M . Nol le , "The A R G U S monitoring system: a reappraisal," in Trends in Computer-Processed Electrocardiograms, J . H. van Bemmel and J . L. Wi l lems (editors). Amsterdam: North-Hol land, 1977, pp. 11 - 19. [82] D. E . Mohrman and L. J . Heller, Cardiovascular Physiology. New York : M c G r a w H i l l , 1986. [83] G . S. Bause et al. , "Sk in avulsion during oscillometry," J. Clin. Mon., vo l . 2, no. 4, pp. 262, 263, 1986. [84] G . Celor ia et al. , "Compartment syndrome in a patient monitored with an automated blood pressure cuff," / . Clin. Mon., vo l . 3, no. 2, pp. 139 -141, 1987. 119 [85] F. K . Forster and D. Turney, "Osci l lometr ic determination of diastolic, mean, and systolic blood pressure - a numerical model," Trans. ASME, J. Biomech. Eng., vo l . 108, pp. 359 - 364, 1986. [86] D. J . Hughs, C . F. Babbs, L. A . Geddes, and J . D. Bour land, "Measurements of Young 's modulus of elasticity of the canine aorta with ultrasound," Ultrasonic Imaging, vo l . 1, pp. 356 - 3 6 7 , 1979. [87] D. Z immerman, M . D . , anesthesiologist, personal communication. [88] I. Guttman and S. S. W i l ks , Introductory Engineering Statistics. New York : W i ley and Sons, 1965, pp. 241 - 2 4 8 . [89] R. Maheswarn, A V . Zezu lka, J . S. G i l l , M . Beevers, P. Davies, and D. G . Beevers, "Cl in ica l evaluation o f the Copa l UA-251 and the Dinamap 1848 automatic blood pressure monitors," /. Med. Eng. and Technol., vo l . 12, no. 4, pp. 160 - 163, 1988. [90] C . S . A . , Electromedical Equipment, C22.2 no. 125-M1984. Ottawa: Canadian Standards Associat ion, 1984. A P P E N D I X Patient Data File 121 ADAPTIVE TOURNIQUET SYSTEM Version 2.1 Copyright 1989 by Mark M i l l e r T r i a l number: 10 A. Hospital: VGH Operating room: LSP OR-20 B . Participating s t a f f : 1. Surgeon(s): Dr. Claridge/Dr. Charles Fisher 2. Anesthesiologist: Dr. Yu 3. Biomedical Engineer(s): M. Mi l l e r C. Patient data: 1. Diagnosis: Avascular necrosis, l e f t talus 2. Type of operation: Biopsy, l e f t talus 3. Location of tourniquet cuff: Left prox. thigh 4. Cuff type: 18 x 100 cm 5. Circumference of limb: 44 cm, tubular 6. Sex: M 7. Age: 37 8. Weight: 57 kg 9. Height: 177 cm Time of case: Tue Aug 22 13:58:20 1989 D. Systems report: Time: Message: L.O.P. Occl. Pr. 0.00 PROGRAM BEGINS 0.00 Con. pr. mode ON 250 mm Hg 0.08 Adp. pr. mode ON 1.88 monitoring... 130 mm Hg 250 mm Hg 2.25 monitoring... 125 mm Hg 250 mm Hg 2.45 monitoring... 121 mm Hg 250 mm Hg 2.68 monitoring... 119 mm Hg 250 mm Hg 2.93 monitoring... 123 mm Hg 250 mm Hg 3.16 monitoring... 122 mm Hg 250 mm Hg 3.58 monitoring... 122 mm Hg 250 mm Hg 3.86 monitoring... 117 mm Hg 2S0 mm Hg 4.25 monitoring... 115 mm Hg 250 mm Hg 4.61 monitoring... 126 mm Hg 250 mm Hg 4.85 monitoring... 126 mm Hg 250 mm Hg 5.03 monitoring... 129 mm Hg 250 mm Hg 5.23 monitoring... 134 mm Hg 250 mm Hg 5.41 monitoring... 138 mm Hg 250 mm Hg 5. 63 monitoring... 138 mm Hg 250 mm Hg 5. 65 Con. pr. mode ON 250 mm Hg 5.88 Adp. pr. mode ON 6.05 monitoring... 132 mm Hg 172 mm Hg 6. 26 monitoring... 121 mm Hg 161 mm Hg 6.53 monitoring... 115 mm Hg 155 mm Hg 6.75 monitoring... 120 mm Hg 160 mm Hg 6.96 monitoring... 125 mm Hg 165 mm Hg 7.18 monitoring... 124 mm Hg 164 mm Hg 7.41 monitoring... 120 mm Hg 160 mm Hg 7.63 monitoring... 120 mm Hg 160 mm Hg 7.85 monitoring... 122 mm Hg 162 mm Hg 8.05 monitoring... 122 mm Hg 162 mm Hg 8.26 monitoring... 127 mm Hg 167 mm Hg 8.46 monitoring... 131 mm Hg 171 mm Hg 8. 68 monitoring... 131 mm Hg 171 mm Hg 8.88 monitoring... 128 mm Hg 166 mm Hg 9.10 monitoring... 118 mm Hg 158 mm Hg 9.36 monitoring... 118 mm Hg 158 mm Hg 9. 60 monitoring... 112 mm Hg 152 mm Hg 9.88 monitoring... 112 mm Hg 152 mm Hg 10.11 monitoring... 127 mm Hg 167 mm Hg 10.33 monitoring... 128 mm Hg 168 mm Hg 10.55 monitoring... 125 mm Hg 165 mm Hg 10. 80 monitoring... 125 mm Hg 165 mm Hg 11.01 monitoring... 125 mm Hg 165 mm Hg 11.23 monitoring... 126 mm Hg 166 mm Hg 11.45 monitoring... 123 mm Hg 163 mm Hg 11.85 monitoring... 121 mm Hg 161 mm Hg 12.08 monitoring... 126 mm Hg 166 mm Hg 12.70 CAUTION 12.90 monitoring... 127 mm Hg 167 mm Hg 13.18 monitoring... 126 mm Hg 166 mm Hg 13.38 monitoring... 127 mm Hg 167 mm Hg 13.60 monitoring... 131 mm Hg 171 mm Hg 13. 90 monitoring... 130 mm Hg 170 mm Hg 14.16 monitoring... 128 mm Hg 168 mm Hg 14.41 monitoring... 147 mm Hg 187 mm Hg 14.63 monitoring... 148 mm Hg 188 mm Hg 14.86 monitoring... 144 mm Hg 184 mm Hg 15.06 monitoring... 135 mm Hg 175 mm Hg 15.30 monitoring... 127 mm Hg 167 mm Hg 15.53 monitoring... 123 mm Hg 163 mm Hg 15.76 monitoring... 124 mm Hg 164 mm Hg 16.11 monitoring... 123 mm Hg 163 mm Hg 16.43 monitoring... 126 mm Hg 166 mm Hg 16.76 monitoring... 126 mm Hg 166 mm Hg 17.01 monitoring... 121 mm Hg 161 mm Hg 17.26 monitoring... 115 mm Hg 155 mm Hg 17.51 monitoring... 109 mm Hg 149 mm Hg 17.76 monitoring... 112 mm Hg 152 mm Hg 17.96 monitoring... 114 mm Hg 154 mm Hg 18.16 monitoring... 115 mm Hg 155 mm Hg 18.43 monitoring... 115 mm Hg 155 mm Hg 18.63 monitoring... 113 mm Hg 153 mm Hg 18.83 monitoring... 116 mm Hg 156 fnm Hg 19.26 monitoring... 118 mm Hg 158 mm Hg 19.46 monitoring... 116 mm Hg 156 mm Hg 19.73 monitoring... 126 mm Hg 166 mm Hg 19.93 monitoring... 126 mm Hg 166 cam Hg 20.13 monitoring... 123 mm Hg 163 mm Hg 20.58 monitoring... 116 mm Hg 156 am Hg 20.86 monitoring... 113 mm Hg 153 tarn Hg 21.08 monitoring... 110 mm Hg 150 mm Hg 21.33 monitoring... 110 mm Hg 150 mm Hg 21.65 monitoring... 105 mm Hg 145 mm Hg 21.91 monitoring... 106 mm Hg 146 mm Hg 22.15 monitoring... 106 mm Hg 146 nm Hg 22.48 monitoring... 108 mm Hg 148 mm Hg 22.75 monitoring... 101 mm Hg 141 ram Hg 23.26 monitoring... 112 mm Hg 152 mm Hg 23.53 monitoring... 107 mm Hg 147 mm Hg 23.85 monitoring... 117 mm Hg 157 mm Hg 24.06 monitoring... 112 mm Hg 152 mm Hg 24.35 monitoring... 106 mm Hg 146 mm Hg 24.61 monitoring... 100 mm Hg 140 mm Hg 24.88 monitoring... 101 mm Hg 141 mm Hg 25.13 monitoring... 105 mm Hg 145 mm Hg 25.33 monitoring... 104 mm Hg 144 mm Hg 25.61 monitoring... 103 mm Hg 143 mm Hg 26.25 monitoring... 98 mm Hg 138 mm Hg 26.51 monitoring... 102 mm Hg 142 mm Hg 26.80 monitoring... 108 mm Hg 148 mm Hg 27.01 monitoring... 111 mm Hg 151 mm Hg 27.30 monitoring... 106 mm Hg 146 mm Hg 27.56 monitoring... 99 mm Hg 139 mm Hg 27.90 monitoring... 97 mm Hg 137 mm Hg 28.38 monitoring... 96 mm Hg 136 mm Hg 28.65 CAUTION 29.05 monitoring... 106 mm Hg 146 mm Hg 29.26 monitoring... 126 mm Hg 166 mm Hg 29.50 monitoring... 122 mm Hg 162 mm Hg 29.73 monitoring... 123 mm Hg 163 mm Hg 29. 95 monitoring... 119 mm Hg 159 mm Hg 30.16 monitoring... 118 mm Hg 158 mm Hg 30.35 monitoring... 114 mm Hg 154 mm Hg 30.55 monitoring... 110 mm Hg 150 mm Hg 30.81 monitoring... 107 mm Hg 147 mm Hg 31.08 monitoring... 97 mm Hg 137 mm Hg 31.36 monitoring... 96 mm Hg 136 mm Hg 31. 63 monitoring... 97 mm Hg 137 mm Hg 31.88 monitoring... 107 mm Hg 147 mm Hg 32.10 monitoring... 110 mm Hg 150 mm Hg 32.56 monitoring... 10B. mm Hg 148 mm Hg 32.60 monitoring. .. 101 mm Hg 141 mm Hg 33.05 monitoring... 103 mm Hg 143 mm Hg 33.26 monitoring. .. 102 mm Hg 142 mm Hg 33.51 monitoring... 105 mm Hg 145 mm Hg 33.75 monitoring... 107 mm Hg 147 mm Hg 34.01 monitoring... 110 mm Hg 150 mm Hg 34.26 monitoring... 114 mm Hg 154 mm Hg 34.48 monitoring... 113 mm Hg 153 mm Hg 34.66 monitoring... 115 mm Hg 155 mm Hg 34.68 monitoring... 114 mm Hg 154 mm Hg 35.06 monitoring... 112 mm Hg 152 mm Hg 35.31 monitoring... 109 mm Hg 149 mm Hg 35.56 monitoring... 105 mm Hg 145 mm Hg 35.83 monitoring... 105 mm Hg 145 mm Hg 35.86 Con. pr. mode ON 250 mm Hg 38.45 PROGRAM TERMINATED Time at end of case: Tue AUQ 22 15:07:04 1989 Comments: -Patient's limbs had l i t t l e fat; observed strong signal. -used image Intenslfler for biopsy. Switching off/on x-ray did not cause any signal glitches or software anomalies. -very good case. Observed LOP's lower than SBP due to smaller limb circumference. -ran system In f u l l adaptive mode for most of case. No bleeding was observed. -no bruises observed when cuff removed. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0065495/manifest

Comment

Related Items