Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Development and initial evaluation of tactile displays and tactile alert schemes for physiological monitoring Ng, Yee Lam Ginna 2008

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

Item Metadata

Download

Media
24-ubc_2009_spring_ng_yee_lam_ginna.pdf [ 2.65MB ]
Metadata
JSON: 24-1.0067059.json
JSON-LD: 24-1.0067059-ld.json
RDF/XML (Pretty): 24-1.0067059-rdf.xml
RDF/JSON: 24-1.0067059-rdf.json
Turtle: 24-1.0067059-turtle.txt
N-Triples: 24-1.0067059-rdf-ntriples.txt
Original Record: 24-1.0067059-source.json
Full Text
24-1.0067059-fulltext.txt
Citation
24-1.0067059.ris

Full Text

Development and Initial Evaluation of Tactile Displays and Tactile Alert Schemes for Physiological Monitoring by  Yee Lam Ginna Ng  BASc., The University of British Columbia, 2005  A THESIS SUBMITTED N PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF Master of Applied Science  in  FACULTY OF GRADUATE STUDIES  (Electrical and Computer Engineering)  The University of British Columbia (Vancouver)  November 2008  © Yee Lam Ginna Ng, 2008  Abstract Current standard monitoring systems typically comprise multiple visual and au ditory displays to convey the physiological status of an anesthetized patient to the attending anesthesiologists during a surgery. The overwhelming amount of infor mation conveyed by these systems appears to overload the anesthesiologists, which degrades their responsiveness to the adverse situations in the patient and may ex pose the patient to higher risk. In this thesis, we propose exploring other sensory modalities, such as the sense of touch, to reduce the visual and auditory burden imposed by the current monitoring systems. Tactile technology was first explored in navigation and orientation applica tions for improving the awareness of pilots and drivers to their surrounding en vironment. Recent medical research also began to investigate this technology for clinical applications such as assisting minimal invasive surgery and needle-based im age guided therapy. No study, however, has examined the use of tactile technology to convey physiological information to the anesthesiologists. We therefore propose the use of a wearable tactile display as a complementary advisory device for the current monitoring systems. The tactile display can convey physiological information to the anesthesiologists in a silent and subtle manner without worsening the noise pollu tion problem in the already noisy operating room. The use of tactile display can help enhance the communication to the anesthesiologists of adverse physiological changes occurred in the patients, thus improving patient’s safety. Four user studies were described in this thesis to evaluate different aspects of II  a wearable tactile display. The first study compared the performance of three tactile display prototypes, namely, an electrotactile display on the forearm, a vibrotactile display on the forearm, and a vibrotactile display on the wrist, We found vibro tactile stimulation to be superior to electrotactile stimulation in terms of training time, alert identification accuracy, and user comfort. Response time and accuracy of vibrotactile alert identification appeared negligibly affected by the choice of stim ulation location on the wrist or the forearm. The second study investigated the potential use of Tactons (structured, abstract tactile messages) in designing a tac tile alert scheme with 36 distinct tactile alerts. We examined the use of rhythm, roughness, and spatial locations of the Tactons to provide users with easily inter pretable alerts and demonstrated an accuracy of 81% in tactile alert identification on the abdomen. A higher accuracy could be achieved by using only rhythm and spatial location. In the third study, we evaluated the tactile perception on the ab domen for four tactile alert schemes distinguished by the two encoding parameters, the mean  ()  and the standard deviation  (0’)  in the number of pulses in the tactile  alerts, Among the four alert schemes tested, results demonstrated an optimal per ception when the scheme with  [i  3, a> 0] was used. A mean accuracy of 94% in  tactile alert identification was achieved for an information transmission rate of 3.73 bits. The final study evaluated the usability of a tactile display on the abdomen under simulated low and high clinical workload conditions. Participants did not en counter difficulties in learning the tactile alerts and using the vibrotactile interface. No statistical differences were detected in the response time and in the accuracy of tactile alert identification between these two workload conditions. This important finding suggests the efficacy of the tactile display in improving the communication to the anesthesiologists on adverse physiological changes occurred in the patients even at high workload condition. Our collective findings demonstrate the possible use of tactile display to enhance physiological monitoring of patients and provide insights for future development of tactile displays in the clinical environment.  111  Contents Abstract  ii  .  Contents  iv  List of Tables  viii  List of Figures  x  Acknowledgements 1  2  xix  Introduction  1  1.1  Motivation  1  1.2  Objective and Scope  2  1.3  Thesis Organization  4  Background 2.1  2.2  7  Overview of the Clinical Environment and the Conventional Clinical Monitoring System  7  2.1.1  Anesthesiologists and Clinical Monitoring Practice  7  2.1.2  Problems with Conventional Alarm Systems  8  Overview of Tactile Technology and Applications  10  2.2.1  Virtual Realty, Tactile, and Forced Feedback  10  2.2.2  Current Tactile Technologies  11  2.2.3  Potential Stimulation Locations  17  iv  2.2.4 2.3  3  The Application of Tactile Technology  Overview of the Somatosensory System and Tactile Receptors  17 .  .  2.3.1  The Somatosensory System  19  2.3.2  Tactile Receptors  21  2.3.3  Mechanoreceptors  22  2.3.4  Tactile Perception  24  User Study #1: Evaluation of Vibrotactile and Electrotactile Stimulation, and Vibrotactile Localization on Forearm and Wrist  30  3.1  Introduction  30  3.2  Methods  31  3.2.1  Tactile Display Prototypes  31  3.2.2  Tactile Display Location  32  3.2.3  Tactile Alert Scheme  32  3.2.4  Confusion Matrix  33  3,2.5  Study Procedures  34  3.3  Statistical Analysis  38  3.4  Results  39  3.5 4  19  3.4.1  Study Participants  39  3.4.2  Training phase  40  3.4.3  Testing phase  41  3.4.4  Questionnaire  44  Discussion  47  User Study #2: Designing a Tactile Alert Scheme with Multi dimensional Tactons  49  4.1  Introduction  49  4.2  Methods  50  4.2.1  50  Tactile Alert Scheme  v  52  4.2.3  Information Transmission Rate  53  4.2.4  Confusion Matrix  53  4.2.5  Study Procedures  54  Statistical Analysis  56  4.4  Results  57  4.4.1  Accuracy  58  4.4.2  Response Time  59  4.4.3  Confusion Matrix  59  Discussion  60  User Study #3: Perception of Rhythm-based Tactile Alerts on the Abdomen  63  5.1  Introduction  63  5.2  Methods  65  5,2.1  Tactile Alert Scheme  65  5.2.2  Tactile Belt Prototype  65  5.2.3  Information Transmission Rate  65  5.2.4  Study Procedures  67  5.3  Statistical Analysis  70  5.4  Results  71  5.5 6  Tactile Display Prototype  4.3  4.5 5  4.2.2  5.4.1  Training Phase  71  5.4.2  Testing Phase  73  Discussion  76  User Study #4: Performance of Tactile Belt Display under Sim ulated Low and High Clinical Workload Conditions  80  6.1  Introduction  80  6.2  Methods  81  vi  6.3  6.4 7  6.2.1  Tactile Display Prototype  81  6.2.2  Tactile Display Location  81  6.2.3  Tactile Alert Scheme  82  6.2.4  Think Aloud Method  83  6.2.5  Workload  83  6.2.6  Confusion Matrix  84  6.2,7  Study Procedures  84  6.2.8  Statistical Analysis  88  Results 6.3.1  Study Participants  6.3.2  Training phase  6.3.3  Testing phase  90  6.3.4  Computer Usability Satisfaction Questionnaire (CUSQ)  93  Discussion  93  Conclusion and Future Work  96  7.1  Limitations  97  7.2  Future Work  98  7.2.1  Evaluation of Tactile Display in the Clinical Environment  7.2.2  Evaluation of Tactile Alert Scheme in the Clinical Environment 99  7.2.3  Exploration of Other Tactile Stimulation Locations  100  7.2.4  Clinical Adoption of the Tactile Display  101  .  .  98  Bibliography  103  Appendix A Statement of Co-authorship  120  vii  List of Tables 3.1  Total number of missed alerts in the testing phase. Percentage ex pressed in brackets represents the percentage of missed alerts over total alerts conveyed. Each alert was conveyed 260 times. I, I, D , 1 and D 2 denotes Increasing Level 1, Increasing Level 2, Decreasing Level 1, and Decreasing Level 2 alerts, respectively  3.2  43  Summarized post hoc results for training time, accuracy, and response time.  Letter “N” denotes the number of measurements analyzed.  The Fisher’s least significant difference (LSD) was used to test the difference between groups.  Statistical differences were detected in  the training time and accuracy. Participants who used EF prototype demonstrated a longer training time and lower accuracy in tactile alert identification, as compared to those who used VF and VW pro totypes. No statistical difference was identified in the response time 3.3  44  Summary of participants’ responses to question statements in the categories of “comfort” (question #1, 3, 5) and “effectiveness” (ques tion #2, 4, 6). Responses were classified as “agree” if responses were between [1  4.1  -  3.5] and as “disagree” for responses between [3.6  -  7].  45  Summary of mean and standard deviation (SD) of accuracy on tac tile alert identification and the corresponding, optimal information transmission rate (ITest) and number of tokens that could be con veyed without error  59  viii  5.1  Maximum likelihood estimation IT rates (ITest) for error-free tactile alert transmission for the four tactile alert schemes. ITe t is expressed 3 in bits and the corresponding tactile alerts in tokens. ITest was de termined by the confusion matrix constructed in the final stage of the study at IT rate  6.1  =  4.32 bits (20 tokens)  76  Summary of ANOVA results in accuracy, missed alerts, and response time in the testing phase. Data are expressed as mean ± SD where appropriate  6.2  90  Performance of participants in the tracking task under LW condition. Participants were required to monitor the change in heart rate (HR) on the simulated anesthesia visual display. The data are expressed as mean ± SD where appropriate  6.3  93  Performance of participants in the tracking task under HW condi tion. Participants were asked to track the changes in heart rate (HR), oxygen saturation (Sp0 ), respiratory rate (RR), and body temper 2 ature (T) on the simulated anesthesia visual display. The data are expressed as mean ± SD where appropriate  6.4  93  Summarized Responses from CUSQ. The responses were ranked on a nominal scale of 1 to 7, with 1 indicating “strongly disagree” and 7 “strongly agree”. The participants could check the “not applicable” (N/A) box if they chose not to respond to a particular statement. “N/A” responses were removed from the analysis. Data are presented 94  as mean (SD) as appropriate  ix  List of Figures 2.1  Location of mechanoreceptors in the skin (top) and the behavior of re ceptors in terms of nerve transmission (bottom). The four mechanore ceptors respond to different vibration frequencies to perceive vibra tions ranging from 0.4 to 500 Hz. The receptors are classified by their structures and their behavior in nerve transmission (modified [42]).  3.1  23  Three Tactile Display Prototypes. The vibrotactile display on the wrist (VW, left), the vibrotactile display on the forearm (VF, middle), and the electrotactile display on the forearm (EF, right). The two stimulation locations were represented by letter A and B  3.2  32  Tactile Alert Scheme with four distinct tactile alerts. The four alerts represented Increasing Level 1 (Ii, top left), Increasing Level 2  (‘2,  bottom left), Decreasing Level 1 (D , top right), and Decreasing Level 1 2 (D , bottom right), respectively. The stimulation of location A was 2 represented by a solid line and location B by a dashed line. Each tactile alert was transmitted twice to ensure the tactile alert could successfully transmit to the participants 3.3  33  Study procedures for one experiment. The 90-minute study was di vided into three experiments. The participants tested only one of the three tactile display prototypes described in each experiment. The study procedures in each experiment were identical  x  35  3.4  Graphical User Interface (GUI) for (a) training and (b) testing phases. In the training phase, the participant clicked the button to learn the four alerts. In the testing phase, the participant identified the conveyed alerts by clicking the appropriate button on the GUI. If the participant failed to identify the alert, the “miss it!” button could be pressed  3.5  37  Definition of response time on tactile alert identification. The re sponse time was defined as the time elapsed between the time when the end of the tactile alert and the instance the participant pressed a button on the GUI  3.6  38  Box-whisker plot comparing the training time between the VF, VW, and EF prototypes. Five possible outliers (labeled as  “+“)  were iden  tified using the 1.5 IQR technique but they were retained in the anal ysis since the data were not normally distributed 3.7  40  Histograms showing the data distributions of training time for VF (left), VW (middle), and EF (right) prototypes. Results from the Shapiro-Wilk test indicated that distributions of data were not nor mal. The data, however, appeared to follow a gamma distribution.  3.8  41  Box-whisker plot comparing the accuracy (left) and response time (right) of tactile alert identification (y-axis) between the VF, VW and EF prototypes (x-axis). The accuracy was defined as the ratio of the correctly identified alerts to the total number of alerts and the response time as the time lag between the activation of the alert and the instance participant responded  xi  42  3.9  Histograms showing the data distributions of accuracy of tactile alerts identification for VF (left), VW (middle), and EF (right) prototypes. Results from the Shapiro-Wilk test indicated that distributions of data were not normal.  However, the data appeared to follow an  extreme value distribution  42  3.10 Modified confusion matrix for VF (top right), VW (bottom left), and EF (bottom right) prototypes. Each state (circle) represents one tactile alert and the path represents the behaviour of participants in identifying the tactile alerts. Missed alerts were excluded from the modified confusion matrix. The tactile alert scheme (top left) is included for reference  45  3.11 Preference of participants for the three tactile display prototypes (VF, VW, EF). The VF prototype was found to be the most favorable tactile display prototype for conveying tactile alerts, whereas the EF prototype was found least favorable 4.1  46  Rhythm parameters of the tactile alert scheme. Three distinct rhythms were used to represent the level of alert a single pulse corresponded  ; middle), and three 2 to Level 1 (Rhi; left), two pulses for Level 2 (Rh pulses for Level 3 (Rh ; right). The duration of activation for each 3 rhythm is illustrated in the figure 4.2  51  Roughness parameter of the tactile alert scheme. Two types of rough ness were used to represent the direction of change in the alert. The modulated, very “rough” signal represented an “increasing” alert  (Roi; left), and an un-modulated, “smooth” signal indicated a “de creasing” alert (Ro ; right) 2  51  xii  4.3  Tactile belt prototype (left) and spatial location of tactile alert scheme (right). Six out of the eight tactors in the tactile belt prototype were used to convey tactile alerts to the users. Each tactor was assigned to represent one physiological event. The event was simplified by numbering the event as Event #1 to #6. The navel was labeled as Event #1(Li) and the other locations were labeled from Event #2 ) in a clockwise direction 6 ) to #6 (L 2 (L  4.4  53  Flow diagram of the study procedures. The study was approximately 60 mm  and included a training phase and a testing phase.  Each  participant tested a tactile belt prototype and the tactile alert scheme of 36 alerts 4.5  55  Graphical user interfaces (GUI) for training phase (left), post-training quiz and testing phase (right). During the training, tactile alerts were conveyed to the participant through the tactile belt prototype when the participant clicked the appropriate buttons on the GUI. In the post-training quiz and the testing phase, the participant used the GUI to identify the tactile alert received. The participant was required to identify, on the GUI, the type, the direction of change, and the level that described tactile alert  4.6  56  Screen shot of Mahjongg Solitaire. Participants were distracted by solving this simple Mahjongg-matching puzzle during the testing phase. Background classical music was generated through headphones worn by the participants to mask the ambient noise from the tactors.  Xl”  .  57  4.7  Accuracy of tactile alert identification.  The leftmost bar demon  strates the overall accuracy of tactile alert identification. The three bars in the center present the individual effects of the Tacton param eters on accuracy. The three bars on the right show the accuracy of the paired Tacton parameters. The error bars refer to the standard deviation 4.8  58  Response time for tactile alert identification.  The bar on the left  illustrates the overall response time of the participants in identify ing the tactile alerts. The two bars on the right demonstrated the response time when the correct (RTcorrect) or incorrect (RTjnc.re) responses were made. A one-way ANOVA detected statistical differ ence between RTc.rect and RTjr€ (p 4.9  <  0.001)  60  Modified confusion matrices for spatial location (top right), roughness (top left), and rhythm (bottom). The state (circle) in the modified confusion matrix represents the component of a Tacton parameter. The path corresponds to the behavior of participants in identifying these components in each of the Tacton parameter. Missed alerts were excluded from the matrix  5.1  61  Logic flow diagram of the anesthesiologist in the presence of an ad verse event in the monitored, anesthetized patient. The tactile dis play prototype plays a role of a supplemental advisory device in this 64  thought process  xiv  5.2  ,o], 41 3 Four tactile alert schemes Si[2,=o], S2[ ,i. and 84[5,1.52j. 3 S3[ ], The tactile alert schemes can be distinguished by the encoding pa rameters, which were the mean  (it)  and standard deviation (u) of the  number of pulses in the tactile alerts of each scheme. The average stimulation time presented at the bottom of each tactile alert scheme represents the mean time required to convey the tactile alert in a particular scheme 5.3  66  Tactile belt prototype (left) and the location of tactile stimulation (middle). The same tactile belt prototype described in Chapter 4 was used in this study. Four C2 tactors (right) located at Li, L2, L3, are L4 were used to convey tactile alerts  5.4  67  Number of tactile alerts required to learn and test in each stage. Each stage included one training phase (Tr(n)) and one testing phase (T (n)), where n denotes the number of tactile alerts conveyed in each stage. A different stimulation location was used in each stage to avoid the unpleasant feeling from the continuous stimulation at a single location. The stimulation location was randomly selected from one of the four designated spatial locations  5.5  69  Number of tactile alerts conveyed in each stage and the correspond ing Information Transmission rate. Participants were required to go through all ii stages in the sequence of n  E  {2, 4, 5, 6, 7, 8, 10, 12, 69  14, i6, 20}tokens 5.6  Screen shot of graphical user interface (GUI) for scheme 82[3,o] with i2 alerts (left) and 20 alerts (right). The same GUI was used in the training and testing phase at each stage  5.7  70  Mean training time (left) and mean number of trials required (right) to learn the tactile alert scheme at each stage. The four tactile alert schemes are represented by the encoding parameter  xv  [t,  o]  72  5.8  Post-Hoc Tukey’s HSD test on the tactile alert scheme for training. The number of trials attempted to learn the tactile alert in each stage was used in the analysis. Statistical difference is illustrated by the star (*) symbol  5.9  73  Mean accuracy (left) and response time (right) for tactile alert iden tification with the four tactile alert schemes. A five-way ANOVA demonstrated a lower accuracy of identifying the tactile alert in Si [2,01, ], 41 30 as compared to the other three schemes (S2[ ,i. and 52 3 S3[ ], ,i, 5 S4[ [). The vertical dashed-line indicates the IT rate at 2.8 bits  74  5.10 Results of post-hoc Tukey’s HSD test on accuracy (left) and response time (right) of tactile alert identification for the four tactile alert schemes. Statistical difference is shown by the star (*) symbol.  .  .  .  75  5.11 The accuracy of tactile alert identification for the four tactile alert schemes below ITthreshold  =  2.8 bits. The two dashed lines represent  the accuracy of tactile alert identification in the range between 99% and 100%  77  5.12 The accuracy of tactile alert identification for the four tactile alert schemes above ITthreshold  =  2.8 bits. The slope and the R 2 value for  each accuracy line are summarized in the legend. The linear approx imation of the curve is considered valid if R 2 > 0.975 6.1  78  Tactile belt display prototype (left) and representation of physiologi cal parameters on the prototype (right). In this study, four out of the six tactors were used to convey vibrotactile alerts. Each of the four tactors represented one physiological parameter: EtCO , pPeak, MV 2 exp, and NIBPmean. The tactile alerts were conveyed to the tactile belt prototype user using BluetoothTM technology  xvi  82  6.2  Tactile alert scheme. The scheme consisted of four distinct tactile alerts used to represent a change in the four physiological parameters. The alerts were generated by varying short (200 ms) and long (1200 ms) vibrating pulses. The changes were described by the direction of change (increasing or decreasing) and level (level 1 and level 2).  6.3  .  83  Flow diagram of the study procedures. The duration of the study was approximately 60 mm  and was divided into a training phase, a  testing phase, and a post-study evaluation phase. Participants in this study were either staff anesthesiologists or anesthesia residents. They were instructed to test the tactile belt prototype under LW and HW conditions 6.4  85  Screen shots of the tracking task in LW (left) and HW (right) condi tion. Participants were asked to perform a tracking task by monitor ing the heart rate in LW condition. They were required to monitor heart rate, oxygen saturation, respiratory rate, and body temperature in HW condition  6.5  87  Bar chart demonstrating the distribution of the required training time for the 29 participants. Four possible outliers, identified using the 1.5 IQR technique and highlighted by the squared box on the bar chart, were not removed from the analysis since the data distribution was non-normal. The mean (± SD) training time was 123.8 ± 71.ls.  6.6  .  89  Accuracy (left) and response time (right) of tactile alert identifica tion under low and high workload conditions. The box-whisker plots depict the upper quartile, median, and lower quartile values (top, middle, and bottom lines of the box), respectively. Possible outliers were identified using the 1.5 IQR technique and marked as =  low workload; HW  =  “+“.  LW  high workload. Outliers were remained in  the analysis since the data were non-normally distributed  xvii  91  6.7  Modified confusion matrix for physiological parameters (spatial lo cation; top) and tactile alerts (rhythm; bottom) for LW and HW conditions. The numbers in the modified confusion matrices corre spond to the number on the reference diagrams of spatial location and tactile alert scheme (left)  92  xviii  Acknowledgements I would like to express my gratitude to the following people: First, I would like to thank my research supervisors, Prof. Guy Dumont and Dr. Mark Ansermino, for their guidance and advice throughout the research project. Special thanks to Dr. Stephan Schwarz for offering his time to answer my many questions, especially on the clinical trial protocols and post-study analysis. Second, I would like to thank my parents and grandmother for their kindness. Without their caring, love and support, I would not be able to complete my program. Third, I would like to thank everyone from the UBC Electrical and Computer Engineering for Medicine Lab (ECEM), especially Dr. Pierre Barralon, for their kind support and assistance. Special thanks to all my friends who have always been around. Last but not least, I would like to express my sincere thanks to all study participants for their time and generous support.  YEE LAM GINNA NG  The University of British Columbia November 2008  xix  Chapter 1  Tntro 1.1  ion  Motivation  The primary objective of clinical monitoring of anesthetized patients in the operating room is to ensure patient safety. Abnormal clinical conditions are currently signaled in one of two ways: 1) by a primitive alarm system automatically triggered when a single physiological parameter fluctuates beyond the preset threshold or 2) by an anesthesiologist visually tracking changes of the physiological signal pattern over time. Human attention is considered the most critical factor in patient safety [68], but the most advanced monitoring system to date cannot guarantee that its results are directly transferred to the anesthesiologists’ cognizance.  This illustrates the  shortcoming in technical advances to reduce adverse outcomes. Demands on human attention have increased significantly with the exponen tial growth in the number of physiological parameters being monitored. The high probability of false auditory alarms have also proven counterintuitive [66] [133] [76]. Clinicians’ visual and auditory senses are subjected to overload; responsiveness to auditory alarms diminishes as the number of alarms increases, and the ability to simultaneously appreciate each physiological parameter on the visual display while  1  maintaining careful observation of the patient is compromised. This cognitive bur den will likely increase with the advent of new monitoring devices unless a new link between the clinicians and the information stream is found.  1.2  Objective and Scope  The objective of this thesis is to develop a tactile display prototype that harnesses a largely underutilized sensory organ, the skin, to convey information from the physiological monitors to the attending anesthesiologists in the operating room. The prototype should convey information that can be readily appreciated without overly distracting the clinicians. The corresponding tactile alert scheme, i.e. structured tactile stimulations that carry information of adverse changes in the physiological parameters of the patients, must be sufficiently sophisticated to encode several levels of unique and readily identifiable alerts. The alert should be communicated to the clinicians in a prompt and effective manner. Instead of replacing the existing visual-auditory alarms, the tactile display is intended to serve as a supplemental clinical advisory device that could augment the awareness of the anesthesiologists on adverse physiological changes occurred in patients. The proposed tactile display will eventually be connected to iAssist [9], a software framework for intelligent patient monitoring, to communicate change point events in physiological trends to the anesthesiologists in the operating room. A tactile alert will be transmitted to the anesthesiologists when an event is identified by the change detection algorithms [136] [137] implemented in iAssist. Details re garding the establishment of the connection between the tactile display and iAssist are outside the scope of this thesis and will not be discussed. Here we will describe the development of tactile display prototypes in full details. The project outlined in this thesis proposes to move tactile technology, first used in the world of virtual reality, into the operating room. Exploring the common human interactions of a “tap on the shoulder”, the tactile display aims to improve 2  the communication to the anesthesiologists on adverse physiological events and to produce a supplemental, practical clinical advisory device. With advanced microsensor and wireless technologies, stimulation of the skin’s sensory receptors through tactile displays offers an innovative and very practical way to re-connect clinicians with their patients: First, tactile communication can provide a subtle signal, rather than outright alarms, to indicate abnormal physiological changes in the patient. It does not detract from other forms of communication or patient interaction nor does it disturb other personnel in the clinical environment. Only the anesthesiologist will receive this information. Second, the sense of touch already forms a crucial keystone in medical diag nosis [92] [131]. Tactile stimulation is a familiar signal within the clinical monitoring environment. It has been combined with inspection, palpation, percussion and aus cultation to form the basis for a wide spectrum of diagnoses. Third, the tactile display does not introduce noise pollution in the operating room and can function well despite the background noise that already exists. Initial experiments indicate that tactile signals compete more successfully for attention than auditory alarms since tactile stimulations are virtually impossible to ignore [89]. Tactile stimulation is not blocked during ancillary tasks even after a protracted period of time. In summary, the tactile display should have the following characteristics: • wearable and unobtrusive to everyday work; • low power consumption, wireless, and light weight; • a pleasant tactile interface that does not cause pain or discomfort to the tactile display users; • demonstrated ability of improving the communication to the clinicians on ad verse physiological changes in the patients by transmitting tactile alerts that 3  are difficult to ignore; • intuitive tactile alerts that facilitate effective information transmission to the attending clinicians.  1.3  Thesis Organization  This thesis describes the development and evaluation of several wearable tactile dis play prototypes and the corresponding tactile alert schemes for physiological mon itoring. We present four user studies to evaluate different aspects of the tactile display. The organization of the thesis is as follows: Chapter 2. Background This chapter is divided into three sections: the first section describes the physio logical monitoring system currently used in the operating room and the problems encountered with the conventional system. The second section gives an overview of existing tactile technologies, including the types of tactile display prototypes and the approach to designing a tactile alert scheme. The third section presents the phys iology of the skin, particularly the types of sensory receptors, their characteristics, and the perception phenomena. Chapter 3. User Study #1: Evaluation of Vibrotactile and Electrotactile Stimulation, and Vibrotactile Localization on Forearm and Wrist The development of three tactile display prototypes, namely, vibrotactile on the forearm, vibrotactile on the wrist, and electrotactile on the forearm, is described. A user study is conducted to evaluate the performance of the three prototypes. These prototypes are then compared in terms of the accuracy and the response time of tactile alert identification. The feedback from the study participants on using different prototypes is also assessed. This chapter further compares the two locations  4  selected to transmit vibration stimulation (wrist and forearm) and evaluates the two methods of tactile stimulation (vibration and electrical). Chapter 4. User Study #2: Designing a Tactile Alert Scheme with Mul tidimensional Tactons The design of a tactile alert scheme of 36 distinct alerts using Tactons [7] is described. Previous research introduced the use of multidimensional Tactons (structured, ab stract vibrotactile messages) in the design of the tactile stimuli and demonstrated an improvement in tactile information transmission. The user study described in this chapter investigates the effectiveness and efficacy of a tacton-based tactile alert scheme in terms of information transmission to the tactile display users. The ab domen is explored as the location for tactile stimulation. Results and implications of the study are addressed. Chapter 5. User Study #3: Perception of Rhythm-based Tactile Alerts on the Abdomen This chapter describes a user study that evaluates the tactile perception of different rhythm-based vibrotactile alert scheme designs. Many studies have been conducted to advance existing tactile technology, but few have investigated the perception of rhythm-based vibrotactile stimuli on the abdomen. Four tactile alert schemes are designed and represented by the encoding parameters  [.t,  u], which are the mean  and the standard deviation of the number of pulses in the tactile alerts of each scheme. This chapter compares the four tactile alert schemes using accuracy and response time of tactile alert identification. The best design in preserving the highest information transmission rate is identified.  5  Chapter 6. User Study #4: Performance of Tactile Belt Display under Simulated Low and High Clinical Workload Conditions A study which assesses the accuracy and response time of tactile alert identification under simulated low-workload and high-workload clinical conditions is described. We have invited certified anesthesiologists and anesthesia residents to participate in the study. Results and implications of the study are described in this chapter. A questionnaire is administered to collect participants’ opinions on the potential use of tactile display in the clinical environment as a supplemental advisory device. Chapter 7. Conclusion and Future Work This chapter summarizes the findings and implications of the four user studies. Lim itations of the studies, future research directions, and related issues are addressed.  6  Chapter 2  Background 2.1  Overview of the Clinical Environment and the Con ventional Clinical Monitoring System  In current clinical practice, anesthesia is administered to patients prior to surgery to prevent any distress and pain the patients would otherwise experience.  The  anesthesiologists are responsible for administering anesthetic drugs and monitoring the condition of the anesthetized patients. This illustrates the important role of the anesthesiologists in ensuring the patient’s safety during the surgical process.  2.1.1  Anesthesiologists and Clinical Monitoring Practice  Weinger et. al. describe the anesthesiologist and the operating room as a complex “human-computer” system whose responsibility is to ensure safe administration of an anesthetic [1271. The anesthesiologist performs technical procedures (e.g. admin ister appropriate drugs) during the surgery and acquires information of the anes thetized patient from various visual and/or auditory displays and other operating room personnel [127]. Guidelines set by the Canadian Anesthesiologists’ Society require the presence of a physician or an anesthesia assistant, under the immediate supervision of an anesthesiologist, throughout the administration of all anesthetics. 7  The use of mechanical and electronics monitor is also recommended to aid vigilance. The auditory and visual alarms for oximetry and capnography are advised not to be indefinitely disabled during the conduct of an anesthetic except during unusual circumstances. This practice would ensure that the anesthesiologist is aware of the physiological changes in the anesthetized patient [107]. Many sophisticated monitoring technologies have been developed to facili tate effective monitoring. The pulse oximetry, for example, is regarded as a gold standard of auditory alarms and is always used for clinical monitoring [117]. The monitoring technologies may differ in forms and/or functions, but they all serve one common purpose  -  to alert the attending anesthesiologist to the presence of con  ditions that would be hazardous to the patient [111]. Modern patient monitoring systems can comprise up to 30 alarms and 70 display indications [133]. The alarm characteristics and signal definitions are regulated by the International Organization for Standardization (ISO) [103].  2.1.2  Problems with Conventional Alarm Systems  Current monitoring systems and alarms demonstrate the ability to alert the anes thesiologists under certain circumstances. However, the increased number of visualaudio alarms appear to introduce alarm and display “pollution” that can impose potential hazards to patient’s safety [133].  Visual Alarms Visual displays can be viewed only from a limited range of positions (directional)  and can be eliminated by readjusting visual attention (optional) [70]. Acquisition of a patient’s physiological information from the visual display would not be possible if the anesthesiologists directed their attention away from the display, thus imposing risk to patient’s safety [89]. Anesthesiologists have found to be less attentive to the visual displays during certain periods of time [100] [76].  8  A study conducted by our research group showed that anesthesiologists have spent only 4% of the total analyzed time in the beginning, middle, and end of the clinical case looking at the visual display [35]. The monitoring performance of anesthesiologists was best when information was acquired solely from the auditory display, mediocre if acquired solely from the visual display, and worst if from the hybrid visual-audio display [104] [98]. Anesthesiologists have demonstrated a much longer time (median of 6s; 90-percentile of 40s) to acknowledge the information presented by the visual display than from the auditory display (median of is; 90thpercentile of 3s) during routine anesthesia in the operating room. This illustrates that the visual display may not be the most appropriate medium to convey urgent information [83]. Auditory Alarms Shortcomings in the conventional auditory alarms also raise concerns in terms of pa tient’s safety. Previous studies have demonstrated that the accuracy of recognizing the auditory alarms within operating room is low. Only 33 to 53.8% of the anes thesiologists, operating room technicians, and operating room nurses can correctly identify the source and the reason of the auditory alarms [34] [77] [82]. The probability of false auditory alarms is high. Research showed that ap proximately 75% of alarms during routine general anesthesia were classified as spuri ous, and only 3% of the auditory alarms contributed to represent patient risk during routine anesthesia [66] [99]. These false alarms were often generated by motion ar tifacts or the light of infrared radiation, such as in the case of pulse oximeters [111]. Anesthesiologists were forced to silence the auditory display as a result of a high occurrence of false alarms [133]. This increases the likelihood of anesthesiologists neglecting important information conveyed by the auditory alarms. Noise pollution is also a concern in the busy operating room. Noise levels in the operating room were found to be higher than the recommended noise level, set  9  by The International Noise Council, in many hospitals worldwide [63]. Contributors of noise pollution problem in the operating room include surgical saws, surgical drills, and, surprisingly, auditory alarms [50] [96]. Noises can impair performance and concentration of the anesthesiologists, reduce their mental efficiency [50] [86], and potentially mask the auditory alarms [82]. Patients are likely to be exposed to a higher risk when the anesthesiologists do not receive critical information from the auditory alarms in a timely manner. These shortcomings in the conventional visual-audio alarm system thus have motivated us to use an alternate sensory organ to convey physiological information to the anesthesiologists.  2.2  Overview of Tactile Technology and Applications  Tactile, originates from the Latin word tactilis meaning “to touch”, describes an object that can be perceived by the sense of touch [1]. Tactile information refers to the information an individual gets through the sense of touch.  2.2.1  Virtual Realty, Tactile, and Forced Feedback  The relationship between tactile technology and virtual reality is first explained.  The United Nations Educational, Scientific and Cultural Organization (UNESCO) defines virtual reality as an immersive and interactive simulation of either realitybased or imaginary images and scenes. Computers and human-computer interfaces are used to connect humans to virtual reality [29] [12]. Such interfaces are supported by hardware and software [109]. Dot matrix printer technologies and Braille systems for the blind have inspired many interfaces to connect humans to the virtual world using the sense of touch. Tactile and force feedback are the two techniques currently used to facilitate the connection between humans and virtual reality through the sense of touch. Tac tile cues are sensations of textures, vibrations, and bumps, whereas force feedback cues are sensations of weight, contours, and motion resistance. Force feedback de 10  vices are capable of stopping the motion of the user, but this function is not found in tactile devices [13]. The term haptic device is generally used to describe devices that facilitate sensations of force feedback and kinesthesia (perception of muscular motion, weight, and position) [13]. The word haptic originates from the Greek word haptikos mean ing “able to touch or grasp” [1]. Details regarding haptic devices are outside the scope of this thesis and will not be discussed.  2.2.2  Current Tactile Technologies  Earlier research has demonstrated the feasibility of incorporating wearable tactile displays onto the human body. Tactile actuators made of piezoelectric actuators, solenoid, electrotactile stimulation, pneumatics, shape memory alloy (SMA), voice coil, thermal feedback, and vibration motors have been integrated into wearable tactile displays.  Piezoelectric Actuators Piezoelectric actuators consist of single or multilayer piezoelectric ceramic materials and are commonly used to drive a tactile matrix (tactile actuators that are arranged into a matrix structure). Piezoelectric actuators use the inverse piezoelectric effect to generate mechanical energy, i.e. tactile stimuli, when they are subjected to electrical energy [80]. These actuators are small in size and have low power consumption. However, the fabrication process of the actuators is complex and expensive due to the demand of precise machining. A high drive voltage (hundreds of volt) is required to obtain a displacement of few millimeters in the actuators [80]. Examples of tactile displays that use piezoelectric actuators include a device developed by Kontarinis et. at. The tactile display has a compliant piezoelectric accelerometer and was applied on finger tips of the users. It was used as vibration sensors for teleoperation (remote operation of a device/machine) [69].  11  Another  vibratory tactile display of image-based textures used a piezoelectric actuator to convey tactile information on the fingertip of users [52]. Pasquero et.  al. used  piezoelectric bimorphs as bending motors in a tactile device, known as STReSS, to produce rapid sequences of tactile images on the users’ finger tip [94]. Later they 2 using the same tactile technology in a developed another device named STReSS Braille display for the blind [73]. Solenoid The solenoid actuator generates tactile stimulation using elastic force that results from the interaction between the magnetized solenoid and the corresponding per manent magnet [138]. Solenoid tactile displays can provide high steady-state force but often have poor power consumption. The solenoid actuator has been used in traditional Braille displays and many other tactile displays. Frisken-Gibson, for example, has developed a four-level finger tip search display for the blind. A virtual image was first translated by a computer into a contour map. The solenoids then pushed up the pins in the device to display the image intensity of the virtual image [37]. Other examples of solenoid tactile dis plays include the haptic pen which comprised a pressure-sensitive stylus and a small solenoid to generate a wide range of tactile stimulations [72]. A wearable sensory aid using a solenoid array has been developed to provide the deaf with tactually encoded information about intonation. The tactile stimulations were applied to the user’s forearm [141]. Khoudja et. al. have also fabricated an interface known as VITAL. This vibrotactile interface with thermal feedback used solenoid-based elec tromagnetic microactuators to transmit tactile information to the fingertip of the users [5].  12  Electrotactile Stimulation The electrotactile display uses an electric current as a stimulus to generate an electric field inside the skin. The nerve activity is induced through anodic and cathodic electrodes that are attached to the user’s skin. Electrodes such as apposed bipolar arrays and concentric arrays are commonly used.  The use of concentric arrays  electrodes is more favorable due to a better confinement to the area of electrotactile stimulation [105]. Kajimoto et. at. describe the approach of electrotactile stimulation using a concept known as the “primary color approach”. This is analogous to the three primary colors for vision, which refers to the combination of three primary colors red, blue, and green  -  -  to produce all colors that humans see. The tactile primary  color approach stimulates each kind of tactile receptor in the skin and combines these stimuli to reconstruct complex tactile sensations [61]. This approach has been applied in the Smart Touch device [62] [59] and other electrotactile display prototypes [60] [58]. Electrodes are often small and light weight, but they may cause discomfort to the users as they are attached directly onto the user’s skin. Users may experience possible invasive impression, sudden pain, or even fear, when they are using the electrotactile display. Furthermore, electrical stimulation may be difficult to confine sensation to a small area, even when electrodes with sufficient density are used. The relationship between the amount of current and generated sensation is unclear and unstable. This explains why electrotactile displays have been developed mainly for research and/or rehabilitation purposes, rather than for consumer use [60]. Pneumatics Pneumatic system generates mechanical motion using pressurized air. The system can take many forms such as air-jets or air-rings. Sata et. at. has developed a data glove that consisted of a non-constrained master arm with a tactile feedback device  13  using air-pressure-based pneumatics. Users can feel the force or the sensation of touch between the slave arm and the virtual object by adjusting the air-pressure inside the pneumatic balloon actuator [101]. Another type of pneumatic system draws air from the suction hole to create a tactile illusion on the skin. For example, a pneumatic tactile alerting system used a simple pneumatic pump to produce a pulsation of frequencies on the automobile driver’s hand to alert the driver of possible problems [31]. Tactile displays for teleoperation have also applied pneumatic systems in their designs [85] [24]. A pneumatic system is light in weight and small in size, but the poor spatial and temporal resolution appear to degrade the performance of the tactile display [125]. Shape Memory Alloy (SMA) A shape memory alloy (SMA) is an alloy that can “remember” its shape when deformed and is capable of returning to its original shape when heated. SMA tactile displays generate a sensation of “vibration” on the skin using the expansion-andcontraction cycle, in which the alloy expands when an electric current passes through it and contracts when cooled down. SMA tactile displays have a good power-to-mass ratio, but often encounter low power efficiency problems during contraction and heat dissipation processes. SMA wires or springs are typically used in tactile displays. These tactile devices are designed mainly for teleoperation and the virtual reality environment [130] [51] [69]. Other examples of SMA tactile displays include a sixty-four element tactile display using SMA wires. Each element of the device consisted of a SMA wire crimped to a sprung pin at one end and a fixed connector at the other end [115]. A 3D form display using coil-form SMA was also developed to display large-scale, dense objects such as human faces or geographical features [79].  14  Voice Coil Voice coils use mini-loudspeakers to supply a fixed frequency and constant amplitude that can be detected by the skin. A voice coil tactile display can convey a moving sensation on the user’s skin over a wide range of speeds [90]. The displacement and force provided by such a device, however, is low [18]. Voice coils are capable of producing only a single frequency and a single amplitude of vibration, and so encoding complex tactile information using this type of tactile display may not be possible [43]. An interface named Active Click was designed for mobile tactile feedback [39]. Other examples include a tactile display used for human telecommunication  and human-machine communication. This device consisted of eight independent speakers that were used to transmit tactile data through vibration. Users can sense the roughness presented by the tactile display through the hand [38].  Another  interface known as TouchMaster was developed to facilitate stimulations of the four fingers and the thumb. The interface consists of voice coil actuators that can convey vibrotactile feedback at a fixed frequency of 210-240 Hz with constant amplitude [16]. Thermal Feedback Thermal tactile displays use Peltier elements to stimulate thermal cues through contact with different materials. Peltier elements adopt the Peltier effect to generate localized heating and cooling stimuli on the skin. When a DC current passes through the circuit, temperature differences are observed at the junctions of two dissimilar conductors. This phenomenon is known as the Peltier effect. Typical thermal tactile displays consist of Peltier elements, thermistors, and a PT (proportional-integral) controller. The controller is used to control the surface temperature of the thermal tactile display [49]. Thermal tactile displays are light weight and small.  Rapid  transmission of tactile information through thermal sensing, however, is restricted  15  due to the poor localization of thermal changes in humans [48]. Developments in thermal tactile displays include the prototype that is com posed of a Peltier module and a PID (proportional-integral-derivative) controller to convey thermal cues associated with finger contact [53]. Another type of thermal tactile display used the Peltier device, PID controller, and the thermal sensor to as sist material identification through finger-object contact [134]. In addition to using conduction, thermal sensing can be achieved without any physical contacts between the device and the user by convection and radiation. Dionisio et. aL, for example, used conduction and convection to facilitate thermal sensing in a glove-like output device called ThermoPad. This device comprised a control system, infrared bulb lamps, small fans, and Peltier elements to improve both the impression of reality and the capability of an individual to orientate oneself in virtual worlds [28]. Vibration Motors A vibrotactile display uses DC motors such as pancake motors or cylindrical motors with an eccentric rotating mass to generate vibration and tactile stimulation on the skin of the user. DC motors have now been produced with significantly reduced power consumption, size, and operation voltage operation (typically 1 to 12V) as a result of extensive development of motors in mobile phones and pagers [116]. However, vibrations of the motor can be irritating to users at times. Many tactile displays have adopted the use of vibration motors. A vibro tactile display named conTACT, for instance, used the tactaid skin transducer to transmit navigation information to the surgeon during surgical interventions [46]. A shoulder pad insert vibrotactile display used DC motors to assist navigation and mo tion guidance and facilitated subtle communication to users [116]. A multi-channel vibrotactile display used the resonator to convey force information during a teleoper ated peg insertion [27]. Other vibrotactile displays include the vibrotactile waist belt that used pager motors to assist waypoint navigation [122] and the low-resolution  16  tactor array that utilized vibrators to facilitate vibrotactile letter reading [135].  2.2.3  Potential Stimulation Locations  Many tactile displays use the finger tip and the palm as the location of tactile stimulation due to the precise tactile sensitivity in these areas [128]. Other body locations, such as torso [118] [123], forearm [129], wrist [8], foot [132], shoulder [116], forehead [58], and tongue [97] [114] [57], have also been explored for tactile stimulation. Many tactile displays typically adopt only one stimulation location and one type of tactile technology to convey tactile information, but the use of multiple stimulation locations [40] and the combined use of multiple types of tactile technologies [4] is not uncommon.  2.2.4  The Application of Tactile Technology  Tactile technology has been expanded from its traditional use in Braille displays to function as orientation, navigation, and communication tools [14]. Recent research has also introduced tactile technology for clinical use.  Orientation Many tactile displays have been developed to provide device users with directional information, such as the Active Belt fabricated by Tsukada et. al. [118]. A cock pit display has also been used in aircraft to assist the pilot in keeping appropriate vertical trajectory. Information on the path angle error was conveyed through the cockpit display to the pilot [91]. A tactile vest has been designed for aeronautic and terrestrial orientation and navigation. It can provide direction, position, and velocity signals to assist in manoeuvring the flight motion stimulator in the de sired direction [140]. Tactile displays have also been used to facilitate search and rescue task in a simulated building-clearing exercise. Directional vibrotactile cues were communicated to alert soldiers to areas of a building that they were currently  17  exposed to, but had not yet cleared [75]. Navigation Another function of the tactile display is to assist users in navigating. Van Erp et. al. have conducted extensive research in the use of tactile displays as navigation tools [120]. They introduced a waypoint navigation tool that can map waypoint direction to vibration location [122]. A tactile display for automobile drivers was also developed [121]. Automobile drivers who used the tactile display demonstrated an improvement in their driving performance and a reduction in mental workload, as compared to drivers who solely used visual cues. Other tactile displays with navi gating functions include the torso-based tactile display that can present navigational cues to human operators [55]. Communication The usefulness of tactile displays depends on the quality of information transmitted to the users. Castle et. at. suggested that information conveyed by tactile displays should be highly intuitive [14]. Brewster et. at. introduced the use of Tactons, or tactile icons, to communicate complex information non-visually [7]. Analogous to icons and earcons in visual and auditory senses, Tactons are structured, abstract tactile messages that can operate spatially and temporally. Attributes such as fre quency, amplitude, waveform, duration, rhythm, and body location have been used in the design of Tactons [7]. A subsequent study tested the perception of rough ness (amplitude modulated sinusoids) and rhythm on the users’ forearm and has demonstrated a high accuracy in identification [10]. Another study demonstrated the possible application of Tactons to present progress information of file downloads from desktop computers to mobile phones [8]. A follow-up study conducted to assess the recognition rate of Tactons in the design of mobile displays. A recognition rate of 81% could be achieved if two-parameter Tactons were used [11].  18  Clinical Tactile technology has been used to facilitate biopsy, ablation, brachytherapy, and needle-based image guided therapy. Yanof et. al. have implemented a tactile feed back system for the needle insertion controller in an integrated CT robot system. The tactile system provided radiologists with positional information of the needle during interventional procedures [139]. A bone reaming system with a temperature tactile sensation display has been introduced to aid internal bone fixation surgery [71]. Tactile technology has also been applied to facilitate minimally invasive pro cedures such as laparoscopy and thoracoscopy. Surgeons who perform minimally invasive procedures may lose the “sense of feel” in exploring tissues and organs within operative sites they would otherwise experience in open surgery. Tactile dis plays can assist surgeons to regain such “feeling” by applying tactile stimulations to their fingers [51] [93] [30]. Prior to the writing of this thesis, our research group conducted initial studies to assess the use of the tactile display as a supplemental advisory device to enhance the situational awareness of anesthesiologists towards adverse conditions. A study conducted in a stimulated operating room demonstrated a faster response time in administering a rescue drug when the tactile display was used alongside standard clinical monitoring [36].  Another study illustrated an improved performance of  users using a tactile display for monitoring alerts, as compared to the conventional auditory display [89].  2.3  Overview of the Somatosensory System and Tactile Receptors  2.3.1  The Somatosensory System  The Somatosensory system is a diverse sensory system that comprises processing centers and receptors to produce sensory modalities of touch, proprioception (body 19  position), and nociception (pain). The cerebral cortex, an outer, grey layer of the brain that covers the cerebrum facilitates the perception of the external environment. The parietal lobe of the cerebral cortex provides perception of touch, pressure, pain, and temperature [42]. The somatosensory system triggers when a sensory neuron, known as a recep tor, responds to a specific stimulus such as mechanical force. Sensation signals are carried by the receptors and travel through the bundles known as peripheral nerves. The signals enter the spinal cord through the dorsal root and then travel through two major pathways, namely, the medial lemniscal pathway and the spinothalamic pathway. The medial lemniscal pathway consists of large fibers that carry signals related to positions of the limbs and preserving touch. The spinothalamic pathway contains small fibers that transmit signals related to pain and temperature [42]. The pathways cross over to the other side of the body in the upward path to the thalamus. Most of the fibers synapse in the ventrolateral nucleus and a few in other thalamic nuclei. Sensation signals originated from the left side of the human body arrive at the thalamus in the right hemisphere of the brain, and vice versa. The signals terminate at the somatosensory receiving area (Si), and possibly the secondary somatosensory cortex (S2), in the parietal lobe of the cortex [42]. The complete transmission process delivers perception of sensation that corresponds to the location where the stimulation occurs. The Skin The skin is the largest human organ which has a surface area of approximately 1. 8m 2 in an average adult [20]. The skin consists of three layers: epidermis, dermis, and hypodermis. The outer skin, called epidermis, is a layer of thick dead skin cells which forms the frontline defense for the immune system. Underneath the epidermis is a middle layer called dermis. Tactile sensory receptors (Section 2.3.2) responsible for the sense of touch are located in epidermis and dermis. The third layer, known  20  as hypodermis, is a layer of fat which includes many blood vessels and nerves. This layer regulates the skin and body temperature in humans [42].  Function of the Skin The skin protects humans from the penetration of bacteria and chemical agents, prevents body fluids from escaping, and regulates body temperature. It also allows humans to sense information from the outside environment.  Description of the  detailed structure and functionality of the skin is outside the scope of this thesis and will not be discssued. Here we will provide an overview of the receptors relevant to tactile sensation, including their characteristics and perception limitations.  2.3.2  Tactile Receptors  Different types of receptors in the somatosensory system facilitate the sense of touch. These receptors are special neurons that can receive information from the environ ment and respond to different kinds of energy [42]. Histological studies have identi fied seven classes of mechanoreceptors, two classes of thermoreceptors, four classes of nocioceptors, and three classes of proprioceptors for perception of mechanical force, temperature, pain, and body position, respectively [21].  These receptors  produce human sensation in pressure, vibration, temperature, electric voltage and current, etc. Mechanoreceptors are of particular interest in the development of tac tile displays because they can be triggered by mechanical forces. The structures and characteristics of different types of mechanoreceptors will be discussed in more detail.  Communication and Information Processing of Receptors Receptors deliver perceptional information to the cerebral cortex through nerve transmission, a process that generates an action potential in the receptors. Sensation information, such as mechanical pressure, is carried and transmitted in the form of  21  an electrical signal.  The pressure-sensitive receptor charges up from its resting  electrical level to a new level upon exertion of a mechanical force. This results in a potential difference known as action potential. The receptor returns to its resting potential within approximately 1 ms.  Frequencies of nerve transmission  differ between receptors [42].  2.3.3  Mechanoreceptors  Located in the epidermis and dermis, four classes of mechanoreceptors, namely, Merkel receptors, Meissner corpuscles, Ruffini endings, and Pacinian corpuscles, are responsible for sensing mechanical stimulations such as pressure and vibration. These mechanoreceptors are classified by their structures and behavior in nerve transmission [42]. Four mechanoreceptors respond to different vibration frequencies ranging from 0.4 to 500 Hz. The absolute sensitivities in these receptors are partially over lapping, which enables simultaneous activation of two or more receptors. Perception can therefore be determined from combined inputs to these mechanoreceptors [6].  Merkel Receptor The Merkel receptor is a disk-shape receptor located near the boundary of epider mis and dermis. The non-encapsulated structure of the receptors allows a sustained response to vibrations with a low frequency of 0.3 to 3Hz and the capability of reflecting fine details [64]. Merkel receptors demonstrate a continuous nerve trans mission in the presence of mechanical pressure, and therefore are classified as slowly adapting type I (SA I) receptor (Figure 2.1) [42].  Meissner Corpuscle The Meissner Corpuscle is situated in the dermis immediately below the epidermis and is the receptor with a stack of flattened cells (Figure 2.1). The receptor has an  22  average size of 80 m by 30 pm in an adult and can be found in the dermal papillae and on the sides of the dermal fingers [106]. The Meissner Corpuscle is responsive to frequency range of 10 to 500 Hz and triggers only at the onset and offset of the stimulation, thus is classified as rapid adapting type I (RA I) receptor [42].  Nerve Transmission  II II JL fi Figure 2.1: Location of mechanoreceptors in the skin (top) and the behavior of receptors in terms of nerve transmission (bottom). The four mechanoreceptors re spond to different vibration frequencies to perceive vibrations ranging from 0.4 to 500 Hz. The receptors are classified by their structures and their behavior in nerve transmission (modified [42]).  Ruffini Ending The Ruffini Ending is a roughly cylindrical capsule that contains many-branched fibers (Figure 2.1). This receptor is classified as a slowly adapting type II (SA II) receptor and is responsive to a frequency range of 15 to 400 Hz and skin stretching [42]. Ruffini Endings exist in joint capsules and play an important role in conveying information about perception occurring in joints. Their location in the human skin, 23  however, has not been confirmed [106]. The Ruffini Ending has been identified on the hairy skin of animals such as cats and monkeys, but no similar association has been reported in humans. The presumed scarcity of Ruffini Ending in human skin implies that the receptor has little contribution to the sensation of touch [106].  Pacinian Corpuscle The Pacinian corpuscle is a layered, onion-shaped capsule that surrounds a nerve fibre (Figure 2.1) [42]. The receptor is considered as the largest corpuscular endings found in the skin [106]. It has an average diameter of 0,5 to 1 mm, with largest possible diameter of up to 1 cm [106]. The Pacinian Corpuscle consists of a thick myelinated fibre which forms a direct connection to the central nervous system. The nerve transmission is observed only at the onset and offset of the mechanical stimuli, and so Pacinian Corpuscle is categorized as the rapid adapting type II (RA II) receptor. The Pacinian Corpuscle is responsive to vibration frequencies between 10 and 500 Hz and is sensitive to the texture of moving fingers [42]. The Pacinian Corpuscle is found in many parts of the body, such as in the deeper part of the dermis in the glabrous (non-hairy) regions, fingers, external genitalia, mammary gland, and the subcutaneous tissue [106].  In summary, Meissner Corpuscle and Pacinian Corpuscle are of most interest in the development of a tactile display due to their rapid adaptation to tactile stimuli and more sensitive response to higher vibration frequencies.  2.3.4  Tactile Perception  Localization of Tactile Stimuli Later chapters in this thesis will present the development of vibrotactile displays on the forearm, wrist, and abdomen, as well as an electrotactile display on the forearm. 24  This section thus focuses on describing the tactile perception of vibrotactile and electrotactile stimuli in these stimulation locations. Vierordt ‘s Law of Mobility states that the closer a location to an anatomical landmark, such as the wrist or elbow, the better the absolute localization of the stimuli. Humans are more readily able to sense the same stimulation pattern at a body position where movement occurs [20].  Cholewiak et.  al.  demonstrated  accuracies of vibrotactile localization on the elbow, wrist, and middle of the arm as 82%, 72%  ,  and 45%  ,  respectively, with tactors (the mechanism used to generate  tactile stimulation) at individual sites spaced 25mm apart [20]. Kim et. al. reported that humans can localize only 2 tactor locations on the dorsal wrist or the volar wrist. A total of 4 locations can be correctly localized on both sides of the wrist [17]. Tactile acuity refers to the sharpness of the sense of touch and is often mea sured by a two-point threshold. The two-point threshold measures the smallest separation that the skin requires to clearly distinguish two point stimuli from a sin gle point. Christman suggested a physical separation threshold of 38.5mm between two adjacent tactors on the forearm for point stimuli to preserve spatial accuracy [22]. Cholewiak et. al. later demonstrated that greater separation between tactor locations was required. The large separation can overcome mechanical and physio logical interactions that would otherwise interfere with the vibrotactile stimulation at the two tactor locations. Increasing the tactor separation from 25mm to 50mm on the forearm improved the accuracy of vibrotactile identification from 46% to 66%. The use of the glabrous (i.e. non-hairy) area of the forearm as the tactile stimulation location is favorable, because the threshold for touch is below the threshold for pain in this area [20]. Vibrotactile localization studies conducted on the abdomen demonstrated a more precise localization on or close to an anatomical anchor point (e.g. the navel and the spine) than on the side of the abdomen. Cholewiak et. al. demonstrated an accuracy of 97% in vibrotactile localization with six stimulation locations including  25  the navel and spine when a pulse of vibrotactile stimuli generated at 250Hz was used. The accuracy decreased when the navel and spine were not selected as the stimulation locations (95%), or when more than six locations were used (74% and 92% for 12 and 8 stimulation locations, respectively). The localization of vibrotactile stimuli varied with the spatial separation between tactors and the location of tactors on the trunk. The vibrotactile localization was found similar in different locations on the abdomen. This suggests a similar range of receptor types at the abdomen [19]. Similar localization experiments have been conducted to evaluate the local ization of electrotactile stimuli on the forearm and fingers.  Effects of electrode  configuration (concentric vs. unifocal) and body axis (longitudinal vs. transverse) were evaluated [47]. Mean errors of localization were 4.1 mm for the unifocal elec trodes and 5.0 mm for the concentric electrodes when the electrotactile stimuli were applied on the forearm using a transversal placement of electrode. Mean localization errors were 21.2 mm for unifocal electrodes and 22.5 mm for the concentric elec trodes when the electrodes were placed longitudinally. The findings suggest that a better electrotactile localization could be achieved when the electrodes are placed transversally. The experiment also demonstrated that the electrotactile stimuli were more likely to localize to the proximal side of the tactor (i.e. electrode). Previous research found that the tactile sensitivity of a skin area varied with the concentration of mechanoreceptors in that area. Comparison of potential stimulation locations detected a higher tactile sensitivity in the finger than the palm [128] [54]. Humans were more readily to identify the vibrotactile messages on the torso than on the forearm [95]. A smaller localization error for electrotactile stimuli was detected on the finger (2.7 mm), as compared to the forearm (13.8mm) [47].  26  Perception of Tactile Stimuli  Previous research in tactile perception has identified numerous perceptual phenom ena in the recognition of the tactile stimuli, such as sensory adaptation, simultaneous stimulation, phantom sensation (summation and saltation), and masking. Sensory adaptation refers to the phenomenon when an individual fails to notice the tactile stimuli after the tactile stimulation has applied to the skin for a prolonged period of time. Hahn et. al. showed that the longer the time the same vibrotactile stimuli is applied to the skin, the more likely the user will adapt to the stimuli and fail to respond to them [44]. Equivalent phenomenon can be detected with electrotactile stimuli.  The adaptation of electrotactile stimuli occurs when  the stimulation is at currents greater than the sensation threshold. Experimental studies showed that the adaptation and recovery phases of electrotactile stimuli are completed in approximately 15 mm  [56].  Another perception phenomenon is observed when the stimulations are gen erated on different parts of the body simultaneously. This phenomenon is analogous to the change blindness in visual sense perception, which is a phenomenon where an individual viewing a visual scene fails to detect large changes in the scene [25]. An experimental study conducted by Tan et. al. used a tactile display to deliver arbitrary sinusoidal waveforms to the thumb, index, and middle finger of the partic ipant’s hand. When multiple fingers were stimulated simultaneously using the same waveform, the study participants failed to determine if the signals were conveyed to two or three fingers. When different waveforms of the same duration were applied to multiple fingers, the participants failed to reliably associate the waveform stim ulated on each finger [112]. This phenomenon was also detected in a study where the participants were asked to detect changes in the tactile patterns presented se quentially on seven locations on the body surface. The patterns consisted of 1-3 vibrotactile stimuli presented for 200 ms with a blank interstimulus interval (1ST) of 800 ms. The study participants were unable to detect a change in the consecutively  27  presented tactile patterns [40]. Tactile masking can be described as an interference of one perceptual stimulus with another causing a decrease or lessening in the effectiveness of perception [32]. The amount of masking of the vibrotactile stimuli varies with three factors: the onset-offset delay of the test and masker stimuli, duration of the masker stimuli, and masker intensity. Remote-site masking between ipsilateral sites on the hand is also possible, but happens only when both masker and test stimulus are within the frequency range of the Pacinian system [126] [124]. A similar masking effect is demonstrated with electrotactile stimuli. A general decline in the masking effect in the fingers is observed as a function of distance and intensity of the electric current. Effect of masking on the little finger appear to be exceptionally strong [119]. Saltation, also known as the cutaneous rabbit, is a type of phantom sensation which describes the shift in the perceived location of a tactile stimulus towards a rapidly delivered subsequent stimulus [67]. This phenomenon was first described by Geldard et. al. as the detection of an abnormal localization of tactile stimuli when widely separated tactile stimulation sites were successively stimulated with a series of taps [41]. The saltation effect varies with the time delay between taps and can be observed with both vibrotactile and electrotactile stimuli. The electrocutaneous rabbit is found to be more vigorous due to a shaper stimulus conveyed by the electrotactile stimuli [41] [110]. The theoretical basis for the cutaneous rabbit effect, however, still remains unknown. Summation refers to the interaction of two point stimuli exhibited as a single phantom sensation [78]. The effect is observed when two tactile stimuli with the same intensity are simultaneously applied to adjacent locations on the skin. The individual would feel a combined sensation midway between the two stimulation locations, instead of feeling the individual stimuli. The summation effect depends on the physical separation of the two stimulation locations, the relative amplitudes, and the temporal order of the tactile stimuli [2].  28  Prior research has established the direction in developing effective tactile devices. We have also taken into account earlier findings and conditions in the operation room to develop the most suitable tactile display for physiological monitoring. This will be described in more details in Chapter 3 to 6.  29  Chapter 3  User Study #1: Evaluation of Vibrotactile and Electrotactile Stimulation, and Vibrotactile Localization on Forearm and Wrist 3.1  Introduction  As the first step in designing a wearable tactile display suitable for our application, we developed three tactile display prototypes, namely, the vibrotactile display on the forearm, the vibrotactile display on the wrist, and the electrotactile display on the forearm. A study was conducted to compare the performance of the three prototypes. As discussed in Chapter 1, the wearable tactile display should be unobtrusive to everyday work of the anesthesiologist, light in weight, able to operate with low  30  power consumption, and capable of communicate tactile alerts to the anesthesiol ogist without causing pain or discomfort to the user. We took into account these design specifications and evaluated the three tactile displays based on the train ing time, the response time and accuracy on tactile alert identification, and the responses of users in using these tactile displays. Comparisons of the method of tactile stimulation (vibrotactile and electrotactile) and the location for vibrotactile stimulation (forearm and wrist) are also discussed.  3.2 3.2.1  Methods Tactile Display Prototypes  We have previously tested a vibrotactile display on the forearm [89]. This study expanded upon the previous forearm-located vibrotactile experiment and compared the vibrotactile display at a different location, the wrist. A different stimulation modality, electrical stimulation, was also applied on the forearm. The electrotactile display on the forearm (EF) used anodic and cathodic currents to selectively stimulate different types of mechanoreceptors in the glabrous skin of the forearm [62]. The constant current, adjustable between 0 to 70 mA, was generated by a low voltage (9 V) nerve stimulator (MicroStim, Neuro Technology, Houston, Texas) using the tetanus mode at 100 pulses/s (Hz). The vibrotactile display on the forearm (VF) and vibrotactile display on the wrist (VW) used DC motors (FM37E, Tokyo Parts Industrial Co., Ltd., Japan) to generate vibrations on the forearm and wrist at a frequency of 140 Hz. Operating at a low voltage of 2.5  -  3,8 V, the motors can provide a large amplitude of vibration  which can be sensed by the human skin.  31  3.2.2  Tactile Display Location  Two locations on the forearm and the wrist were designated for tactile stimulation. The choice of tactor locations was based on findings on tactile perception (Section 2.3.4). The tactile display was worn on the user’s non-dominant hand to minimize the obtrusiveness of the prototype to the user. The three tactile display prototypes (VF, EF, and VW) are shown in Figure 3.1.  — —  —  —  —  LoatnA L:’n  B  Figure 3.1: Three Tactile Display Prototypes. The vibrotactile display on the wrist (VW, left), the vibrotactile display on the forearm (VF, middle), and the elec trotactile display on the forearm (EF, right). The two stimulation locations were represented by letter A and B.  3.2.3  Tactile Alert Scheme  Information was conveyed from the tactile display prototype to the user using the tactile stimulus. Each stimulus was assigned a meaning and regarded as a tactile alert. Four tactile alerts were created by combining long and short pulses (Figure 3.2). These pulses were generated by either electrical or vibration stimulation at ap proximately 100 Hz or 140 Hz, respectively. The tactile alerts were used to represent a change in the level of heart rate of a patient. The four tactile alerts were classified into two groups, with one group corre sponded to an increasing alert and another as the decreasing alert (Figure 3.2). The two tactile alerts in each group were further categorized as a Level 1 and Level 2 alert. The tactile alert with one long pulse and one short pulse was the Level 1 alert, whereas the alert with one long pulse and three short pulses was the Level 2 alert. Level 1 corresponded to a 10% change in observed heart rate of the patient over the 32  previous 5 seconds, and Level 2 referred to a 20% change. In summary, the four tactile alerts were: Increasing Level 1 (Ii), Increasing Level 2  (‘2),  Decreasing Level  1 (Di), and Decreasing Level 2 (D ). The direction of change in the alert (increasing 2 or decreasing) was identified by the sequence of tactile stimulation in location A and B (Figure 3.1). The alert was an increasing alert if location A was stimulated first, and was a decreasing alert if location B was first stimulated. Each tactile alert was transmitted twice, with a separation of 1000 ms between repetitions. The repetition was used to ensure the tactile alert was successfully transmitted to the participants. On  I  2QOrns  D  J1_JL e  e  a  ririri  Level 2  I  _j  :: :: :  I  I  •e  j’’’’’ 4aDm  Levell  e  2C  a  2OOm  Level 2 4:O—  2OOT  2Ofl5  ibme  me LocaUon A Location B  Figure 3.2: Tactile Alert Scheme with four distinct tactile alerts. The four alerts represented Increasing Level 1 (Ii, top left), Increasing Level 2 (12, bottom left), Decreasing Level 1 (D , top right), and Decreasing Level 2 (D 1 , bottom right), 2 respectively. The stimulation of location A was represented by a solid line and location B by a dashed line. Each tactile alert was transmitted twice to ensure the tactile alert could successfully transmit to the participants.  3.2.4  Confusion Matrix  The confusion matrix [15] was used to evaluate the number of tactile alerts that could be sent and correctly identified by the participants.  A first order ri x n  confusion matrix, with n as the number of conveyed tactile alerts, was constructed by specifying row i of the matrix as the conveyed tactile alerts and column j as  33  the alerts identified. The diagonal entries of the matrix thus represented the sum of tactile alerts that were correctly identified. The off-diagonal entries provided the number of tactile alerts that were identified incorrectly. Three confusion matrices, one for each of the three tactile display prototypes, were constructed in this study.  We modified the confusion matrix with a state  diagram to facilitate better representation. The tactile alerts were represented by a state and the response to identify the alerts by the path. Incorrect responses were illustrated by a path traveling from the state of the conveyed alerts to the state of the responded alerts, whereas the correct responses were symbolized by a path returning to its own state. The missed tactile alerts were excluded from the confusion matrix and were analyzed separately.  3.2.5  Study Procedures  This study was approved by the Clinical Research Ethics Board of The University of British Columbia (Board Approval Number: H05-70250). Participants in the study were not medically trained and were recruited through advertisements posted on notice boards in the University. The duration of the study was approximately 90 mm and participants were compensated with a CAD$5.00 coffee shop gift card. The study was divided into three experiments, with each experiment focused on testing one of the three tactile display prototypes described. The order of the sections was randomized using the Matlab (Mathworks Inc.) function ran4perm to reduce bias. Figure 3.3 illustrated the study procedures of one experiment. The procedures for the three experiments were identical. Each experiment included four phases: calibrating, training, testing, and debriefing. In the calibration phase, one tactile display prototype was first worn on the non-dominant hand of the participant. We then adjusted the intensity of the tactile stimulation by conveying some test tactile alerts through the prototype. The purpose of the calibration was to ensure the tactile stimulation would not cause  34  Calibrating Phase  Training Phase  No No .Terminate experiment after 30 mm  Testing Phase  Proceed to Next  End of Study  Figure 3.3: Study procedures for one experiment. The 90-minute study was divided into three experiments. The participants tested only one of the three tactile display prototypes described in each experiment. The study procedures in each experiment were identical. discomfort or pain to the participants; in addition, to ensure the participants could clearly distinguish the alerts conveyed at each of the two locations. During the training phase, the participant was given time to familiarize them selves with the tactile display prototype, the tactile alerts and the corresponding meanings. Using the graphical user interface (GUI), the participant received a tac tile alert by pressing one of the four buttons on the GUI (Figure 3.4(a)).  The  participant then proceeded to take a post-training quiz, where the participant was required to identify the tactile alerts conveyed in random order. The post-training quiz served to determine if the participant has learned most, if not all, of the tactile alerts. The training phase terminated when the participant obtained an accuracy of 80% in identifying the tactile alerts in the post-training quiz. The participant was asked to return to the training if the accuracy of 80% was not achieved. The training phase terminated when the participant obtained the required accuracy in the post-training quiz. The experiment was terminated if the participant failed to achieve the required accuracy after a training period of 30 mm. 35  In the 15 mm  testing phase, the four tactile alerts were conveyed to the  participant via the tactile display prototype in random order and at random time. The participant was required to identify the alerts using the GUI shown in Figure 3.4(b). It was not necessary for the participant to wait until the end of the tactile alert transmission before responding on the GUI. The “miss it” button should be pressed if the participant failed to identify the alert. An alert was also regarded as “missed” if the participant did not press any buttons on the GUI before the next alert was sent. Each of the four tactile alerts (Ii, 12, D , D 1 ) were conveyed 260 2 times in the study. We distracted the participants by simulating an operating room environment. The participant was required to monitor a simulated visual display, which was identical to the one that used in the operating room, and reported any changes in the values of heart rate and systolic blood pressure. Background noise, taken from an operating room during a real surgical procedure, was played through the headphone worn by the participant. The participant then proceeded to the debriefing phase to complete a section of the questionnaire corresponding to the prototype just tested. The questionnaire was administrated to measure the user’s perception of each tactile display prototype and the tactile stimulation locations. The questionnaire consisted of eight questions in total and was divided into three categories: the first category evaluated the user’s comfort in using the three prototypes (“comfort”: question #1, 3, and 5); the second category assessed the level of difficulties in learning the tactile alert scheme using different prototypes (“effectiveness”: question #2, 4, and 6); and the third category collected user’s preference on the method to convey tactile alerts (electrical or vibration) and locations for tactile stimulation (wrist or forearm) (“preference”: question #7 and 8). At the end of each experiment, the participant responded to the questions in categories “comfort” and “effectiveness” by ranking on a nominal scale of 1 to 7. Responses between [1 between [3.6  -  -  3.5] was classified as “agree” and responses  7] as “disagree”.  36  The participant then continued with a further experiment to test a different tactile prototype. At the end of the study, the participant was asked to complete the questions in catergory “preference” in the questionnaire. The participant chose the preferred method of tactile stimulation and ranked the preference of the three tactile display prototype using a nominal scale of 1 (most favorable) to 3 (least favorable). General comments from the participants were also elicited. Alan,,  he1oeargl  -  • Alarm Fejnp  —  Fss k of t hutt,s  Ivy the el-t fe.eli Of Heart  Please dick on the button representing the pattern that detects.  tti f -f HF{ clWes (?tfl tWCt *  e. .L S r,**ett o e o’’  t  LewI 1: 1UZCbage  -J Pee  Level 2: 20% CliIiqe  , the m. ** le hee*.1  I  eva!  Increasing  I 5mw Ciwmaga  Decreasing  -  tavaI2 iaAliqe Increaslnq  Decreasinq  Level I  LOvel I  Level  Levl2  LeveII]  Level i  Level 2  Level 2  Miss It’  EXIT  Figure 3.4: Graphical User Interface (GUI) for (a) training and (b) testing phases. In the training phase, the participant clicked the button to learn the four alerts. In the testing phase, the participant identified the conveyed alerts by clicking the appropriate button on the GUI. If the participant failed to identify the alert, the “miss it!” button could be pressed. For each experiment of the study, data were recorded in each phase except the calibrating phase. The number of trials required to memorize a tactile alert and the duration of the training phase were recorded. The accuracy and the response time for tactile alert identification, as well as the number of missed alerts in the testing phase were also obtained. The accuracy was defined as the ratio of the number of correctly identified alerts to the total number of alerts conveyed. The response time was defined as the elapsed time between the time when the tactile  37  alert was conveyed and the instance the participant pressed a button on the GUI (Figure 3.5). The responses to the questionnaire, as well as demographic data such as age and gender were also collected. -o  fl [i[1[i  —j  I  fl[H1  111111  111111  LILI’_______  r  Response Time  Figure 3.5: Definition of response time on tactile alert identification. The response time was defined as the time elapsed between the time when the end of the tactile alert and the instance the participant pressed a button on the GUI.  3.3  Statistical Analysis  The predicted sample size was determined from the results of a pilot study that compared the response time of tactile alert identification between VF and EF pro totypes. We conducted apriori power analysis targeted to detect a 20% difference in the response time between the two prototypes at a  =  0.05 and 3  =  0.2. A sample  size of at least 20 participants would be required. We applied ANOVA to test for the differences between the three prototypes (VW, VF, and EF) in terms of accuracy and response time to tactile alert identification. Differences were considered significant at p < 0.05. The choice of whether to use parametric or non-parametric statistical test was based on the normality of the data. The Shapiro-Wilk test was used to determine the normality. The data distribution was classified as non-normal at p < 0.05. Parametric statistical testing was used under the following circumstances:  • when the distribution of the data was normal; • when the data distribution departed from normal, yet when ANOVA was 38  applied, the obtained Type I error did not deviate from the designed value (i.e. p < 0.05). Prior studies [65] [23] [102] [113] using Monte Carlo simulation have demonstrated that ANOVA is quite robust against certain non-normal distribution such as gamma distribution and extreme value distribution. For data following these distributions, ANOVA was used. • when experimental designs were complex.  Non-parametric tests are found  most useful for one-way designs and are much less so for complex experimental designs [65]. Outliers identified by the 1.5 inter-quartile-range (IQR) technique [108] were re moved from the statistical analysis only if the data was normally distributed. This was to prevent outliers from skewing the results in a set of normally distributed data [84]. Yet, for non-normally distributed data, we did not remove possible outliers to avoid discarding representative data from a long-tailed, non-normal distribution. Results from the questionnaire were calculated by averaging all responses collected from the participants. The analysis was performed using Matlab (Mathworks Inc., Natick, MA, USA).  3.4 3.4.1  Results Study Participants  The 30 participants in this study tested the VF, VW, and EF prototypes. Four participants were excluded from the analysis due to technical failure of the EF pro totype. The analysis and conclusions of the study were based on the data collected from the remaining 26 participants (9 females and 17 males with 70% of participants belonging to the 20-29 yr age group).  39  Training phase  3.4.2  A total of 78 measurements (26 participants x 3 prototypes) were recorded. Only 76 measurements were analyzed, since two measurements (one for VF and one for VW) were accidentally erased. The data tested with the Shapiro-Wilk test were found to be non-normally distributed (p  =  0.001 for VF, p  =  0.005 for VW, and p  =  0.027  for EF), but the data appeared to follow a gamma distribution (Figure 3.7). A total of 5 possible outliers were detected (Figure 3.6) but they were not removed from the analysis, since the data distributions were non-normal. 500 C’,  400 Co  +  p300 +  200 *  I  —.-  a) 100 E I-  0  HVF  H VW  EF  Prototype  Figure 3.6: Box-whisker plot comparing the training time between the VF, VW,  and EF prototypes. Five possible outliers (labeled as “+“) were identified using the 1.5 IQR technique but they were retained in the analysis since the data were not normally distributed. A one-way analysis of variance (ANOVA) revealed a significant difference [F(2, 73)  =  3.15,p  =  0.049] between groups. A post-hoc analysis of the difference  between groups was then tested using Fisher’s least significant difference (LSD) =  (  0.05). Participants required a longer training time to become familiar with the  EF prototype (Table 3.2, column 3).  40  t  0 11)  2 z  Training Time (s)  Figure 3.7: Histograms showing the data distributions of training time for VF (left), VW (middle), and EF (right) prototypes. Results from the Shapiro-Wilk test indi cated that distributions of data were not normal. The data, however, appeared to follow a gamma distribution.  3.4.3  Testing phase  Accuracy A total of 78 measurements (26 participants x 3 prototypes) were analyzed for accuracy. The data were non-normally distributed (Shapiro-Wilk test, p VF, p  =  0.004 for VW, and p  =  =  0.012 for  0.002 for EF), but appeared to follow the extreme  value distribution (Figure 3.9). We detected 7 possible outliers (Figure 3.8) but they were retained in the analysis since data distributions were non-normal. The accuracy for VF, VW, and EF were 94.7%, 94.3% and 88.8%, respectively. A one-way ANOVA revealed a significant difference between groups [F(2, 75) = 0.025]. The LSD post hoc analysis (c  =  =  0.05) was used to test for differ  ences between groups, and indicated that the EF prototype was less accurate than the VF and VW prototypes (Table 3.2, column 5).  41  100  H  H—  l  7000  90  6000 + +  >‘  80  D C.) C.)  70  C.)  1  +  I I  + +  5000 4000  L +  3000  60  1-  +  VW  VF  I  EF  VF  EF  Prototype  +  Prototype  Figure 3.8: Box-whisker plot comparing the accuracy (left) and response time (right) of tactile alert identification (y-axis) between the VF, VW and EF prototypes (x axis). The accuracy was defined as the ratio of the correctly identified alerts to the total number of alerts and the response time as the time lag between the activation of the alert and the instance participant responded.  IC  € 0 5 5 0  3  2  40  60  60  160 Accuracy  (%)  Figure 3.9: Histograms showing the data distributions of accuracy of tactile alerts identification for VF (left), VW (middle), and EF (right) prototypes. Results from the Shapiro-Wilk test indicated that distributions of data were not normal. However, the data appeared to follow an extreme value distribution.  42  Prototype VF VW EF  I  12  2 D  2 (0.8%) 3 (1.2%) 9 (3.5%)  0 (0.0%) 3 (1.2%) 10 (3.9%)  4 (1.5%) 6 (2.3%) 18 (6.9%)  6 (2.3%) 4 (1.5%) 11(4.2%)  Mean (SD) 3.0 (1.3) 4.0 (1.1) 12.0 (3.7)  Table 3.1: Total number of missed alerts in the testing phase. Percentage expressed in brackets represents the percentage of missed alerts over total alerts conveyed. 2 denotes Increasing Level 1, Each alert was conveyed 260 times. Ii, 12, D , and D 1 Increasing Level 2, Decreasing Level 1, and Decreasing Level 2 alerts, respectively. Missed Alerts Table 3.1 summarizes the total number of missed tactile alerts for each of the four alerts. The percentage in bracket represents the percentage of missed alerts with respect to the total alerts conveyed (i.e. 260 alerts). A two-way ANOVA detected no significant difference between the number of missed alerts on the four tactile alerts, but did demonstrated a statistical difference between tactile prototypes [F(2,300)  =  6.89,p  =  0.0012j.  A LSD test further  suggested that participants tended to miss a higher amount of tactile alerts when the EF prototype was used, as compared to the VF and VW prototypes. Response Time A total of 78 measurements (26 participants x 3 prototype) for the response time (RT) were obtained. The data were normally distributed (Shapiro-Wilk test, p 0.14 for VF, p  =  0.44 for VW, and p  =  =  0.38 for EF). Two outliers were removed  from the analysis (Figure 3.8). The mean response times were 5.63 s, 5.53 s and 5.60 s for the VF, VW, and EF prototypes, respectively. A one-way ANOVA detected no statistically different between groups [F(2, 73) 7).  43  =  O.l4,p  =  0.87] (Table 3.2, column  Prototypes VF VW EF  Training Time N Means (s) 63.4 25 61.0 26 113.6 25  N 26 26 26  Accuracy Means (%) 94.7 94.3 88.8  Response Time N Means (s) 25 5.63 26 5,53 25 5.60  Table 3.2: Summarized post hoc results for training time, accuracy, and response time. Letter “N” denotes the number of measurements analyzed. The Fisher’s least significant difference (LSD) was used to test the difference between groups. Statistical differences were detected in the training time and accuracy. Participants who used EF prototype demonstrated a longer training time and lower accuracy in tactile alert identification, as compared to those who used VF and VW prototypes. No statistical difference was identified in the response time.  Confusion Matrix Figure 3.10 illustrates the modified confusion matrix for the VF, VW, and EF prototypes. Participants, using the VF prototype, demonstrated more difficulties , and D 2 2 , 12 from I or D 1 , D 2 1 from I or D in distinguishing I from 12 or D from D 1 or 12. A similar phenomenon was observed in the other two prototypes. Participants experienced higher confusion in identifying the tactile alerts using the EF prototype, and encountered more difficulties in determining the direction of change (i.e. increasing or decreasing) of the tactile alerts using the VW prototype.  3.4.4  Questionnaire  A total of 26 measurements (26 participants x 1 questionnaire) were collected. The question statements and responses of participants in the categories “comfort” and “effectiveness” were summarized in Table 3.3. Responses obtained from the “comfort” category showed that over 90% of participants considered the VF and VW prototypes comfortable to use. In contrast, only 58% of the participants regarded the EF prototype as comfortable. A one-way ANOVA analysis confirmed that the EF prototype was less comfortable to con  44  0, IC I,  ‘2  02  JJJU1IL  Figure 3.10: Modified confusion matrix for VF (top right), VW (bottom left), and EF (bottom right) prototypes. Each state (circle) represents one tactile alert and the path represents the behaviour of participants in identifying the tactile alerts. Missed alerts were excluded from the modified confusion matrix. The tactile alert scheme (top left) is included for reference.  Statement 1. The VF prototype did not cause any discomfort. 2. The tactile alerts were easy to learn using the VP prototype. 3. The VW prototype did not cause any discomfort. 4. The tactile alerts were easy to learn using the VW prototype. 5. The VF prototype did not cause any discomfort, 6. The tactile alerts were easy to learn using the EF display.  Agree 96.2% 88.5%  Disagree 3.8% 11.5%  92.3% 92.3%  7.7% 7.7%  57.7% 73.1%  42.3% 26.9%  Table 3.3: Summary of participants’ responses to question statements in the cat egories of “comfort” (question #1, 3, 5) and “effectiveness” (question #2, 4, 6). Responses were classified as “agree” if responses were between [1 3.5] and as “dis agree” for responses between [3.6 7]. -  -  45  vey tactile alerts than the other two vibrotactile display prototypes (VF and VW) [F(2,75)  =  8.68, p < 0.001]. Feedback from the “effectiveness” category demon  strated that 88.5%, 92.3%, and 73.1% of the participants reflected that the tactile alerts conveyed by the VF, VW, and EF prototypes, respectively, can be recognized easily. An one-way ANOVA detected no significant difference between the three prototypes [F(2,75)  =  2.28, p  =  0.111. Responses from the “preference” category  demonstrated that 80.8% of the participants favored the use of vibration as the method to convey tactile alerts. Only 19.2% preferred electrical stimulation. Fur thermore, 46.2% of the participants chose the VP prototype as their most favorable tactile display prototype, 42.3% for the VW prototypes, and 11.5% for the EF pro totypes. Over 50% of the participants did not favor the use of the EF prototype to receive tactile alerts (Figure 3.11), 100  —w 90 EF 70  ? 60 50 40  P1 ii if1 First  Second Choice  Third  Figure 3.11: Preference of participants for the three tactile display prototypes (VF, VW, EF). The VF prototype was found to be the most favorable tactile display  prototype for conveying tactile alerts, whereas the EF prototype was found least favorable.  46  3.5  Discussion  In the pursuit of developing a tactile display that would be of use to anesthesiologists in the operating room for patient monitoring, our study indicates that, even when administered to a group of non-experts, a vibrotactile display was able to effectively communicate tactile alerts in a manner that was superior to the electrotactile display. With regard to the location of the vibrotactile display, we detected no dif ference in the response time to identify the tactile alerts between the forearm or the wrist. The purpose of recording response time was, in part, to evaluate which location, the forearm or the wrist, resulted in a more rapid identification of an alert and, based on this criterion the two vibrotactile display locations were equivalent. Pertaining to the confusion matrix, participants demonstrated a higher con fusion between the directions of change (increasing or decreasing) or between the levels (level 1 or level 2) of tactile alerts. Recalling that the direction of change in the tactile alerts was distinguished by the sequence of tactile stimulation in loca tion A and B, the findings suggested that participants may experience difficulties in memorizing the sequence of stimulation. Also, since the levels of the tactile alerts bear the same number of long pulses and only differ by the number of short pulses, participants were more likely to confuse a tactile alert with another when they failed to detect the exact number of short pulses conveyed. Appraising the confusion matrix with the accuracy of tactile alert identifica tion, the participants experienced higher confusion in identifying the tactile alerts when the EF prototype was used. Some participants reported in the informal poststudy discussion that the EF prototype failed to transmit a sharp tactile stimulus. This may provide the rationale of the higher confusion and lower accuracy encoun tered by the participants when they used the EF prototype to identify the tactile alerts. The observation was supported by the findings in the questionnaire that over 80% of the participants preferred the use of vibrotactile stimulation to convey the alerts. 47  From a usability perspective, the electrotactile display failed to transmit a sharp tactile alert to the user and hence introduced more confusion in alert identifica tion. We observed a longer calibration time in adjusting the intensity of stimulation in the EF prototype, as compared to the VF and VW prototypes. The use of nonreusable electrodes was inconvenient and costly. Electrodes were attached to the user’s skin and were likely to cause discomfort to the users during use and at the time when the electrodes were removed from skin. Based on the above considerations, we concluded that the vibrotactile display prototypes (VF or VW) yielded superior communication of information, due to their ability to convey accurate tactile alerts efficiently and effectively, whilst being both user-friendly and comfortable. No statistical difference was found between the accuracy and response time of using the wrist or the forearm on the user’s nondominant hand to receive vibrotactile alerts. End-user preference could therefore be permitted in choosing either location for the clinical vibrotactile alert system. Further investigation is required to design a more effective tactile alert scheme so as to reduce the level of confusion in identifying different tactile alerts. The improved tactile alert scheme should avoid using the sequence of stimulation on different locations as a factor to distinguish alerts. A more sophisticated tactile alert scheme is also required to capture all the necessary physiological information for future applications.  48  Chapter 4  User Study #2: Designing a Tactile Alert Scheme with Multidimensional Tactons 4.1  Introduction  The findings in Chapter 3 demonstrated the potential use of vibrotactile stimulation to convey tactile alerts to the tactile display users. The results also revealed the need of a more sophisticated alert scheme to transmit the complex array of physiological information available in current monitoring systems. This chapter evaluates a complex tactile alert scheme of 36 distinct alerts designed with Tacton parameters [7]. The abdomen is explored as an alternative location, for tactile stimulation, apart from the forearm and the wrist investigated in Chapter 3. Using a commercially available vibrotactile belt prototype to con vey tactile alerts, we conducted a study to assess the effectiveness of the tactile alert scheme. The optimal number of tactile alerts that can be conveyed by the vibrotactile belt prototype was evaluated using the modified confusion matrix. The Tacton parameters required to preserve a high degree of accuracy for tactile alert 49  identification was identified.  4.2 4.2.1  Methods Tactile Alert Scheme  Tactons are structured, abstract tactile messages that can operate spatially and temporally (Section 2.2.4). Research conducted by Brown et. al. utilized Tacton parameters of rhythm, roughness, and spatial location to convey tactile stimuli on the forearm of the tactile display users and demonstrated an overall accuracy of 71% in tactile stimuli recognition [10]. We expanded these findings and designed a similar tactile alert scheme for the abdomen by varying the Tacton parameters (rhythm, roughness, and spatial location). We collectively employed three types of rhythm, two types of roughness, and six spatial locations on the abdomen to create a tactile alert scheme of 36 distinct alerts.  Rhythm (Rh) Adapting the concept of musical notes and rests, rhythm represents a combination of different durations of pulses separated by distinct pauses. Previous experiments using the forearm to convey tactile information found an accuracy of 96% in stimulus recognition when three types of rhythm were used [11]. Based on these findings, we employed three distinct rhythms in the design of the alert scheme. Each type of rhythm represented a distinct alert level: a single pulse corresponded to Level 1  ) 3 ), two pulses for Level 2 (Rh 1 (Rh ), and three pulses represented Level 3 (Rh 2 (Figure 4.1).  Roughness (Ro) Roughness refers to the amplitude modulation of the signal created by multiplying a sine wave of one frequency with the sine wave of another frequency. Brown et al.  50  400 ms  200  1200 mc  800 mc  ms  1000 mc  Figure 4.1: Rhythm parameters of the tactile alert scheme. Three distinct rhythms ; 1 were used to represent the level of alert a single pulse corresponded to Level 1 (Rh ; right). 3 left), two pulses for Level 2 (Rh ; middle), and three pulses for Level 3 (Rh 2 The duration of activation for each rhythm is illustrated in the figure. [10] demonstrated the successful use of roughness to convey tactile information on  the forearm. Based on these observations and the presumption that skin perception between the forearm and the abdomen were similar, two types of roughness were used in our design to represent the direction of change in the alert (Figure 4.2). A very “rough” signal was represented by a 250 Hz sine wave modulated by a 30 Hz sine wave, and “smooth” signal by a non-modulated sine wave generated at 250 Hz. The very “rough” signal indicated an “increasing” alert (Roi) and the “smooth” signal corresponded to a “decreasing” alert (Ro ). 2  3;,  3;3  EW7W  1W  a  7W  Figure 4.2: Roughness parameter of the tactile alert scheme. Two types of roughness were used to represent the direction of change in the alert. The modulated, very “rough” signal represented an “increasing” alert (Roi; left), and an un-modulated, “smooth” signal indicated a “decreasing” alert (Ro ; right). 2  51  Spatial Location (L) Accurate localization of a tactile stimulus can be used to impart further detailed information. Earlier experiments on the localization of vibrotactile stimuli on the abdomen demonstrated that the tactile localization is most accurate when the stim ulus is close to an anatomical reference point (i.e. the navel or the spine). When a stimulus was generated at 250 Hz at six different locations around the abdomen, an accuracy of 97% in tactile stimulus localization was observed. The accuracy de creased when more than six locations were used (Section 2.3.4). We thus adapted the use of six spatial locations on the abdomen, with each location corresponding to one distinct physiological event (Figure 4.3), Since the focus of the study was to evaluate human perception of the tactile alert scheme, the assignment of the six physiolog ical events was simplified by numbering each event as Event #1 (Li) to Event #6 ). The location at the navel was numbered as Event #1 and the subsequent 6 (L events were labeled as #2 to #6 in clockwise direction. The simplified assignment would reduce potential difficulties for the non-anesthesiologist participants (Section 4.2.5), who generally lacked knowledge that would make the physiological events meaningful.  4.2.2  Tactile Display Prototype  A tactile belt prototype manufactured by Engineering Acoustics (Winter Park, FA) was used (Figure 4.3). The belt prototype consisted of eight C2 tactors (electrome chanical vibrotactile transducers) that were separated an equal distance apart. Six out of the eight tactors were used in this study. The tactors were arranged in a dorsal and ventral cluster of 3 tactors each. Referencing the navel as the 12 o’clock position, the tactors at the 3 o’clock and 9 o’clock positions were not used.  52  Figure 4.3: Tactile belt prototype (left) and spatial location of tactile alert scheme (right). Six out of the eight tactors in the tactile belt prototype were used to convey tactile alerts to the users. Each tactor was assigned to represent one physiological event. The event was simplified by numbering the event as Event #1 to #6. The navel was labeled as Event #1 (L ) and the other locations were labeled from Event 1 ) to #6 (L 2 ) in a clockwise direction. 6 #2 (L  4.2.3  Information Transmission Rate  Information transmission (IT) rate is a measure of the correlation between the amount of information encoded and the amount of information received [112]. The number of tactile alerts communicated to participants in this study was expressed in terms of IT rate, with the corresponding number of alerts reported as tokens. For a particular alert-response pair (Si, Ri), IT can be described as  IT  =  P (S/R) P(S)  l0g2  where P(S/R) is the probability of the conveyed alert S given the response R, and P(S) the a priori probability of (Si). The average IT (ITavg) is given by the  weighted sum of log [P(S/R)/P(S)]: k I  avg  k i,  =  j)092  j=1 i=1  P(S/R) F(S) ‘I  The information transmitted was measured in bits.  4.2.4  Confusion Matrix  The number of tactile alerts that could be sent and identified by the participants can be evaluated using the modified confusion matrix presented in Chapter 3. The 53  maximum amount of information that the tactile belt prototype can transmit with out error (i.e. 100% accuracy), denoted as ITest, can be computed by approximating the probabilities by the frequency of occurrence: ITest  =  -lO92  (mii.n)  j=1 i=1  where n is the total number of trials in the experiment, rijj the number of times the joint event (Si, R) occurs, and ri  =  nj and nj  =  nj the row and  column sums, respectively.  4.2.5  Study Procedures  The study was approved by the Clinical Research Ethics Board of The University of British Columbia (Board Approval Number: H07-00012). Non-medically trained participants were recruited from the general public in Vancouver, B.C., through on-campus and online advertisement at the University. Each participant received CAD$10 as an honorarium for participation. The duration of the study was approximately 60 mill and included a training phase and a testing phase (Figure 4.4). Upon written consent, the participant was given a training session to become familiar with the vibrotactile belt prototype and the tactile alert scheme of 36 distinct alerts. The participant received tactile alerts by clicking the buttons on the graphical user interface (GUI) at their own pace (Figure 4.5). The participant then proceeded to take a post-training quiz to assess their ability in identifying the tactile alerts. We transmitted each of the 36 tactile alerts once and the participant was required to identify the alert by choosing the type of alert event, the level of the event, and the direction of change in the alert on the GUI. A minimum accuracy of 60% in the post-training quiz was required to proceed to the testing phase; otherwise, the participant was asked to return to the training. The training phase terminated when the participant achieved the minimum accuracy in the post-training quiz, or the study was terminated if the participant  54  failed to achieve the minimum accuracy in the quiz after 30 mm of training. Training Phase  No Testing Phase  Terminate section after 30 mm  End of Test  Figure 4.4: Flow diagram of the study procedures. The study was approximately 60 mm and included a training phase and a testing phase. Each participant tested a tactile belt prototype and the tactile alert scheme of 36 alerts. During the testing phase, the participant was asked to identify 72 tactile alerts (36 alerts conveyed twice) conveyed in random order and at random time within the 30 mm  test. The participant identified the information encoded in the  tactile alert on the same GUI used in the post-training quiz. The participant was required to identify (1) the level of alert (Level 1, 2, or 3), (2) the type of event (Event #1 to #6), and (3) the direction of change (increasing or decreasing). The participant was able to choose to respond on the GUI without necessarily waited un til the end of transmission of the tactile alert. The participant was given an interval of 30s to identify each tactile alert. The alert would be regarded as “missed” if no response was made on the GUI 30s after the tactile alert was conveyed. Through out the study, the participant was distracted by playing a simple puzzle game called Mahjongg Solitaire (Figure 4.6). Background classical music was also played through the headphone to mask the ambient noise generated by the tactors of the prototype. Data on the accuracy and response time of the tactile alert identification were collected in the testing phase. The accuracy of tactile alert identification was defined as a ratio of correctly identified alerts to the total number of alerts conveyed,  55  whereas the response time for tactile alert identification was defined as the interval between the activation of the alert and the instant the participant clicked the button on the GUI. :  -  Event ‘  ‘LJ  rJ L..02  lraoasng va,  .tJ L&d2  Event#l  ‘  Event #2  ‘  Eventf3  Ev,ot#4  ‘  Event#5  ‘  Ev,nt#6  ____j  1  Type of A#avn  -  r Doaoosnn -  vvn,v  easg  ‘hl  -  a’  Level otAlaen  -tz  ‘  Level 1  -  ‘  Level 2  Level’  Boek to MOAn  1  Figure 4.5: Graphical user interfaces (GUI) for training phase (left), post-training quiz and testing phase (right). During the training, tactile alerts were conveyed to the participant through the tactile belt prototype when the participant clicked the appropriate buttons on the GUI. In the post-training quiz and the testing phase, the participant used the GUI to identify the tactile alert received. The participant was required to identify, on the GUI, the type, the direction of change, and the level that described tactile alert.  4.3  Statistical Analysis  Sample size calculation was based on prior studies [11] [19], in which a minimum size of 20 participants was required. Analysis of variance (ANOVA) was used to test for statistical differences between the three Tacton parameters (spatial location, roughness, and rhythm) in terms of the accuracy of tactile alert identification. We also applied ANOVA to test for statistical differences between the response time to make correct responses (RTorred) and incorrect responses (RTjnct ’,.rect) in tactile 7 alert identification. Differences were considered significant at p  <  0.05. Tukey’s  HSD test was used for post-hoc comparisons. The choice of whether to use nonparametric or parametric statistical test followed the criteria presented in Section 56  Mthjon  Figure 4.6: Screen shot of Mahjongg Solitaire. Participants were distracted by solving this simple Mahjongg-matching puzzle during the testing phase, Background classical music was generated through headphones worn by the participants to mask the ambient noise from the tactors. 3.3. We anticipated the presence of extreme cases, and therefore, all data were included in the analysis. A modified confusion matrix was constructed for each Tacton parameter (rhythm, roughness, and spatial location) to determine ITest. Analysis was performed using Matlab (Mathworks Inc., Natick, MA, USA).  4.4  Results  A total of 30 non-medically trained participants (16 females and 14 males) were recruited from The University of British Columbia. Each participant completed the study with sufficient data generated to determine the accuracy and response time of tactile alert identification. Due to a technical failure in the tactile belt prototype, we were unable to identify the reason of “missed” alerts (i.e. due to a technical failure or the participants failing to identify the alerts). The “missed” alerts were therefore excluded from the analysis.  57  4.4.1  Accuracy  An overall accuracy of 81.3% with a standard deviation (SD) of 6.4% was found in tactile alert identification (Figure 4.7). We found that the roughness parame ter gave the lowest accuracy (mean: 88.7%, SD: 9.8%) and rhythm parameter the highest accuracy (mean: 96.3%, SD: 3.3%). The accuracy for the spatial location parameter was 95.1% (SD: 9.3%) (Table 4.4.1). We also evaluated the accuracy of the paired Tacton parameters. The highest accuracy was found in the locationrhythm pair (mean: 91.6%, SD: 9.5%). The accuracy for the roughness-rhythm pair and the location-roughness pair were 85.7% (SD: 11.6%) and 84.4% (SD: 12.7%), respectively. 110 100  go 80 70 60 0  4o 30 20 10 0 q,  c.  Figure 4.7: Accuracy of tactile alert identification. The leftmost bar demonstrates the overall accuracy of tactile alert identification. The three bars in the center present the individual effects of the Tacton parameters on accuracy. The three bars on the right show the accuracy of the paired Tacton parameters. The error bars refer to the standard deviation. A three-way ANOVA was conducted to compare the effect of each Tacton parameter on accuracy. A significant effect of roughness [F and rhythm [F  =  4.4’7,p  =  =  0.49,p 4  <  0.0011  0.01] on the accuracy of tactile alert identification 58  was detected. A Tukey-Kramer honestly significant difference (HSD) test further revealed that the Level 1 rhythm Rh 1 (Figure 4.1) was more efficient in information transfer than Level 3 Rh , whereas roughness Ro 3 1 (“rough”) demonstrated a higher accuracy than Ro 2 (“smooth”). The roughness-spatial pair also demonstrated a significant effect on the accuracy [F  =  2.42,p  =  0.033].  Tactons  Rh  L  Ro  L & Rh  Ro & Rh  L & Ro  %correct SD  81.3 6.4  96.3 3.3  95.1 9.3  88.7 9.8  91.6 9.5  85.7 11.6  84.4 12.7  ITest (bits) tokens  4.18 18.07  1.32 2.49  2.24 4.71  0.50 1.41  3.60 12.11  1.83 3.56  2.76 6.78  Table 4.1: Summary of mean and standard deviation (SD) of accuracy on tactile alert identification and the corresponding, optimal information transmission rate (ITest) and number of tokens that could be conveyed without error  4.4.2  Response Time  The overall mean response time (RT) for tactile alerts identification was 4.8 s (SD: 2.5 s).  Response time of the participants in making correct and incorrect  responses was also determined. The response time for correct tactile alert identi fication (RTcorrect) was 4.7 s (SD: 2.3 s) and 5.5 s (SD: 3.0 s) for incorrect tactile alert identification (RTjncorrect) (Figure 4.8). The distributions were non-normal (Shapiro-Wilk test: p  <  0.001 for the distributions of RTc ,.rect and RTjnc,,.re). A 0  Kruskal-Wallis ANOVA demonstrated a significant difference between RTcrect and RTincorre  2 {x  =  38.11, p  <  0.001], in which the response time for making a correct  response was faster.  4.4.3  Confusion Matrix  The modified confusion matrices for the three Tacton parameters (spatial location, roughness, and rhythm) are presented in Figure 4.9. The modified confusion matrix of the spatial location revealed that participants encountered more difficulties in distinguishing the tactile alert localization between neighboring locations. A higher 59  1 p<o.oQ 5  I  0  Correct reponse Incorrect response  Figure 4.8: Response time for tactile alert identification. The bar on the left ii lustrates the overall response time of the participants in identifying the tactile alerts. The two bars on the right demonstrated the response time when the cor rect (RTcorrect) or incorrect (RTincorrect) responses were made. A one-way ANOVA detected statistical difference between RTcorrect and RTjncorrect (p < 0.001). confusion rate was detected in the ventral cluster than in the dorsal cluster. Partic ipants were more likely to confuse Ro 2 (“smooth”) with Ro 1 (“rough”) when using , 1 the roughness parameter. In contrast, the level of confusion between rhythms Rh , and Rh 2 Rh 3 were similar. An overall ITest of 4.18 of information (18.07 tokens) can be conveyed when three types of rhythm, two types of roughness, and six spatial locations were used (Table 4.4.1). In addition, the individual ITest for each Tacton parameter was 1.32 bits (2.49 tokens), 0.50 bits (1.41 tokens), and 2.24 bits (4.71 tokens) for rhythm, roughness, and spatial location, respectively.  The location-rhythm pair demon  strated a highest ITest of 3.60 bits (12.11 tokens).  4.5  Discussion  The tactile belt prototype worn on the abdomen exhibited an optimal ITest of 4.18 bits (18.07 tokens) with no error in tactile alert identification by varying two types of rhythm, one type of roughness, and four spatial locations. Further investigation 60  NAVEL  s4 Roughness  Spatial Location  “-4 .SPNE  JilL Ri1IL f.1  Rhythm  Figure 4.9: Modified confusion matrices for spatial location (top right), roughness (top left), and rhythm (bottom). The state (circle) in the modified confusion ma trix represents the component of a Tacton parameter. The path corresponds to the behavior of participants in identifying these components in each of the Tacton parameter. Missed alerts were excluded from the matrix.  61  ], [Roi], and [L 2 , 2 in the modified confusion matrices suggested the use of [Rh , Rh 1 , L 3 L , L 4 ] for roughness, rhythm, and spatial location, respectively. 5 Comparing these findings to previous work [11], the performance in distin guishing roughness on the abdomen (88.7%) was higher than that on the forearm (82.4%). The “rough” alert (Roi) was more easily recognized than the “smooth” alert (Roi). Participants in this study demonstrated a lower overall accuracy in iden tifying roughness parameter (below 90%) than recognizing the other two parameters (rhythm and spatial location, above 95%). The tactile perception of roughness on the abdomen was relatively poor and cannot be considered the optimum candidate for our tactile alert design. Spatial location proved to be the parameter that exhibited the best overall accuracy, but a significant interaction between spatial location and roughness was detected. This thus confirms that the roughness-location pair is likely to have a negative effect on accuracy and should be avoided. Participants spent more time in responding when they were uncertain about which tactile alert they had received and gave a delayed incorrect response. This result reinforces the common assumption that more time is required in responding to uncertain situations. This study has provided insight to the possible use of Tacton parameters in the design of a complex tactile alert scheme to encapsulate sophisticated physio logical information. The findings further suggest using rhythm and spatial location in the design. In-depth investigation of the systematic approach of varying these two Tacton parameters is needed to facilitate unambiguous delivery of useful infor mation to the participants in the experiment, and conceivably would do so for an anesthesiologist in an operating room setting.  62  Chapter 5  User Study #3: Perception of Rhythm-based Tactile Alerts on the Abdomen 5.1  Introduction  Chapter 4 demonstrated the potential use of Tactons in the design of a complex tactile alert scheme. The results suggested varying rhythm and spatial location to preserve high accuracy in tactile alert identification. This chapter continues with these findings and explores the appropriate approach to design an effective rhythmbased tactile alert scheme suitable for identification of alerts around the abdomen. To derive the suitable tactile alert scheme, knowledge of how the anesthesiolo gist extracts information from the tactile display prototype should first be explained (Figure 5.1). The physiological monitoring unit activates the tactile display in the case of an adverse event in the anesthetized patient (Step 1). The signal received from the physiological monitoring unit is then mapped to the corresponding tactile alert (Step 2). The tactile display prototype then transmits the tactile alert to the attending 63  I —;--‘  ( Physiological )  Activation  ZPI  coding Conversion from  Tactile display is  pyog  prese of an adverse event  0  TransmIssion Tactile alert is conveyed through thectlledlsplay  () Decoding  Reception  Conversion from tacile alert into physiological event  Tactile alert is detected by skin receptors (Pacinian, lleissner, Merkel, Ruffini)  Anesthesiologist  Figure 5.1: Logic flow diagram of the anesthesiologist in the presence of an adverse event in the monitored, anesthetized patient. The tactile display prototype plays a role of a supplemental advisory device in this thought process. anesthesiologist (Step 3). The anesthesiologist detects the tactile alert on the skin (Step 4), decodes the information embedded in the alert, and provides the necessary treatment to the patient (Step 5). The effectiveness of the tactile alert scheme thus depends heavily on how a physiological event is encoded in a tactile alert and how much information can be en capsulated in the alert scheme to preserve the accuracy in tactile alert identification. This chapter addresses these research problems by investigating the human tactile perception with regard to different tactile alert scheme designs. We introduced four distinct rhythm-based tactile alert schemes, each consisting of 20 alerts, and evalu ated the accuracy and response time to tactile alert identification as the number of transmitted tactile alerts increased. We aim to identify the most appropriate tactile alert scheme design that would preserve high accuracy and rapid response time.  64  5.2 5.2.1  Methods Tactile Alert Scheme  Four distinct tactile alert schemes, each consisted of twenty different tactile alerts, were designed (Figure 5.2).  All tactile alerts were presented with varying short  (200 ms) and/or long (600 ms) pulses separated by a time interval of 200 ms, 400 ms, 800 ms, 1400 ms, 2000 ms, or 2600 ms. We classified each alert scheme using the encoding parameters. The encoding parameters were the mean  ()  and the  standard deviation (u) of the number of pulses in the tactile alerts of each scheme. We described the mean number of pulses  (t)  in each tactile alert scheme as,  =  and the standard deviation (u) of the number of pulses in the tactile alert scheme as, =  (‘-) (N  )2)1/2  where N referred the number of pulses in the  jth  alert, and N 0 the total number  of alerts included in the tactile alert schemes..  5.2.2  Tactile Belt Prototype  We used the same tactile belt prototype described in Chapter 4 (Figure 5.3). The prototype was worn on the abdomen and contained eight C2 tactors (electrome chanical vibrotactile transducer). In this study, only four of the eight tactors were used. Referencing the navel as 12 o’clock position, tactors at 12 o’clock, 3 o’clock, 6 o’clock and 9 o’clock positions were used.  5.2.3  Information Transmission Rate  As introduced in Chapter 4 Section 4.2.3, we expressed the number of tactile alerts communicated to participants in terms of Information Transmission (IT) rate. The maximum likelihood estimation of error-free Information Transmission  65  Stimulus Scheme Si  P1fl  P2J1J P4J]_[  Stimulus Scheme S2 ; u=O  JLflfl flflfl JLJU1  flrL JUl JLfl J1 IlfiJi  JUL  JUUI  JUUUl  j1J1J :flJIJ]  P9 P10  JLfl JLfl 1LH JLfI  P11 P12 P13  P14 P15 P16 P17 P18  P20  fl fL fl fl flfl flfl flfl flJ] f1f[  Stimulus Scheme 54 cr=152 j5  ‘li iU1i fliUl fUlfill iuuLrl. fliUlil. JUUUI JIIIflJ1 IIDJU1  JUu JIJ1T[ LflJ  FLf1J1.  P6  Stimulus Scheme 53 u=T..41 u= 3  ‘fl Hi. H1JI j1fl  flllfl JU11 JfflLfl flfl J1JUIJ1. j1J1f1j  flJ I1J flJ’fl flJLJI JJfl flJJ flJLfl flfl j1fl JUU]1  lflJLflft J1JIIUWL ff11flJl JjJJ JU1JJ1J1JI flf]JJ1j1  J1jJU1JS1Ji. J1JJ1JJ1J  ijJJl J1JUU1JLflJU1  Average  stimulation time  2200 ms  2160 ms  1580 ms  2540 ms  ,o], 4 3 Figure 5.2: Four tactile alert schemes Sl[=2,=o], S2[ ,i. 3 S3[ i ], and S4[s,i. ]. 52 The tactile alert schemes can be distinguished by the encoding parameters, which were the mean () and standard deviation (u) of the number of pulses in the tactile alerts of each scheme. The average stimulation time presented at the bottom of each tactile alert scheme represents the mean time required to convey the tactile alert in a particular scheme.  66  LI flveh  L4  12  •  C2 Tactor  Figure 5.3: Tactile belt prototype (left) and the location of tactile stimulation (mid dle). The same tactile belt prototype described in Chapter 4 was used in this study. Four C2 tactors (right) located at Li, L2, L3, are L4 were used to convey tactile alerts. rate (ITest) was evaluated using the approach outlined in Chapter 4 Section 4.2.4. Both IT and ITest were measured in bits with the corresponding number of alerts reported as tokens.  5.2.4  Study Procedures  This study was approved by the Clinical Research Ethics Board of The University of British Columbia (Board Approval Number: H07-00688). Study participants were not medically trained and were recruited through online advertisements posted on the University’s intranet. The participants were compensated with CAD$iO as an honorarium. The duration of the study was approximately 90 mm  and each participant  ], 82[3,o], 20 was randomly assigned to test one of the four tactile alert schemes (Si[ 1521 using the tactile belt prototype. , 15 S4 i. or ) 41 , 3 S3[ ],  The assignment of the  schemes was generated using the Matlab (Mathworks Inc.) function rand. The study had ii stages. Each stage was further divided into a training phase (Tr(n)) and a testing phase (T(n)) (Figure 5.4). All participants were required to go through each stage chronologically to learn and test a group of tactile alerts. The number of alerts conveyed (n) in each stage increased sequentially: n  {2,4,5, 6, 7, 8, iO, i2, 14, 16,20}tokens  67  This corresponds to the IT rate of IT  =  1092(Th)  =  {1.00, 2.00,2.32,2.59,2.81,3.00,3.32,3.59,3.81,4.00, 4.32}bits  Table 5.5 summarizes the number of tactile alerts and the corresponding IT rate conveyed at each stage. During the training phase Tr(n), the participants were asked to familiarize themselves with a set of tactile alerts, Tr(n)Vn e {2, 4, 5,6, 7,8, 10, 12, 14, 16, 20}tokens by clicking the appropriate button on the graphical user interface (GUI) in Figure 5.6. Participants learned 2 alerts in Tr(2), 4 alerts in Tr(4), 5 alerts in Tr(5), and so on. A random spatial location on the abdomen (Li, L2, L3, or L4) (Figure 5.3) was chosen at the beginning of each training phase. This was to avoid the unpleasant sensation from continuous stimulation at a single location. The training phase Tr(n) terminated when the participants indicated that they had remembered the tactile alerts. They would then proceeded to the testing phase. We measured the training time and number of trials attempted to learn the tactile alerts in each training phase. In the testing phase, the participants were required to identify the same set of tactile alerts they learned in the preceding training phase. All n tactile alerts were conveyed randomly on four occurrences to each participant. Therefore, each participant received a total of 416 tactile alerts in the testing phase. Using the same GUI as in the training phase (Figure 5.6), the participants identified the conveyed tactile alerts by clicking the appropriate button on the GUI. The spatial location used for tactile stimulation in the testing phase T(n) was the same as that used in the training phase Tr(n). Participants were required to listen to classical music using the headphone. This was to mask the noise generated by the tactile belt prototype. Measurements  68  on the accuracy and response time of tactile alert identification were collected dur ing the testing phase. The accuracy was defined as the ratio of number of alerts identified correctly to the total number of alerts transmitted. The response time was the time elapsed from the end of the tactile alert transmission to the time the  participant clicked a button on the GUI. Due to same technical failure of the tac tile belt prototype described in Chapter 4, missed alerts (i.e. alerts that were not identified by the participants) were excluded from the analysis. Information on the gender, age, and years of formal musical training of the participants were collected. Tr(2) T(2)  Tr(5) T(5)  Tr(4) T(4)  -———-  Tr(i) T(i)  --—---  Tr(20) T(20)  ++ Figure 5.4: Number of tactile alerts required to learn and test in each stage. Each stage included one training phase (Tr(n)) and one testing phase (T(n)), where n denotes the number of tactile alerts conveyed in each stage. A different stimulation location was used in each stage to avoid the unpleasant feeling from the continuous stimulation at a single location. The stimulation location was randomly selected from one of the four designated spatial locations.  Number of pattern In) IT (bits)  1 0  3 1.59  Numberofpattern(n) IT (bits)  11 3.46  13 3.70  —  ...2  6 2.59  15 3.91  L4.OQJ  16  9 3.17  .  2.81 17 4.09  18 4.17  19 4.25  Figure 5.5: Number of tactile alerts conveyed in each stage and the corresponding Information Transmission rate. Participants were required to go through all 11 stages in the sequence of ii {2, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20}tokens Miller et. al. found that human can remember a maximum of 7 different patterns at one time [81]. We therefore expected that participants should demon strate a high accuracy in identifying up to 7 tactile alerts (i.e. ITthreshold  =  2.8  bits), regardless of type of scheme used. We hypothesized that participants would attain an accuracy of tactile alert identification between 99% and 100% when the IT rate was below 2.8 bits. We also hypothesized that the accuracy would decay linearly when IT rate was increased above 2.8 bits. 69  pU..  ruu-1  ifi  LL.J  flJ j  flU nruj  n_nj  LJJI  IL_U  Jifli  JWlfl  flj  TLiLS1  FlU  fl_U  flTLfl  LILJ,  nnfluninsj  j  Figure 5.6: Screen shot of graphical user interface (GUI) for scheme S2[3o] with 12 alerts (left) and 20 alerts (right). The same GUI was used in the training and testing phase at each stage.  Statistical Analysis  5.3  The sample size calculation for this study is based on the previous results on accuracy of tactile alert identification using the vibrotactile display on the forearm (Chapter 3).  An a priori power analysis was used to hypothesize a 5% difference in the  accuracy of tactile alert identification between two tactile alert schemes at c and 3  =  =  0.05  0.2. A sample size of at least 15 participants was calculated as being  required to test each of the four tactile alert schemes (Si , S2 1201 01 S3[3,i.4i] and , 13 S4[5,1.52}). Thus, a total sample size of 60 participants was required. A five-way analysis of variance (ANOVA) was used to test for statistical dif ferences between IT rate at each stage (IT  =  3.59, 3.81, 4.00, 4.32} bits), tactile alert scheme  {1.00, 2.00, 2.32, 2.59, 2.81, 3.00, 3.32, ,o], S3[3,1.41] S4[5,1.52]), 3 [2o] S2[  gender (male, female), musical training (yes, no), and age (below the median age, or above) in terms of the accuracy and response time of tactile alert identification. Differences were considered significant at p < 0.05 and the Tukey’s HSD test was used for post-hoc comparisons. The choice of whether to use non-parametric or parametric statistical test followed the criteria described in Section 3.3. Measure ments recorded at the stage of IT rate  =  1.00 bit (2 tokens) were excluded from the  analysis. Participants used the training and testing phases at IT rate 70  =  1.00 bit to  familiarize themselves with the GUI and tasks they were required to perform. All data were included in the analysis as we anticipated the presence of extreme cases in identifying the tactile alerts. The accuracy of tactile alert identification was assessed around ITthreshold =  2.8 bits. Linear regression was applied to the accuracy slope of the four tactile  alert schemes above 2.8 bits. A valid linear approximation was indicated by an  2 > 0.975. All analysis were performed using Matlab (Mathworks Inc., Natick, R MA, USA).  5.4  Results  A total of 64 participants (37 male and 27 female) were recruited from The Uni versity of British Columbia. All participants were either undergraduate or graduate students from the University. The age range was between 18 to 41 yr, with a me dian age of 24 yr. Among these participants, 35 had formal music training and 29 no music background. All participants were randomly assigned to one of the four ], 20 tactile alert schemes. The number of participants was 15, 16, 16, and 17 for S1[ ,o], 41 3 S2[ ,i. and ], 3 S3[ ], 52 respectively. All participants completed all stages of . 1 , 5 S4[ the study.  5.4.1  Training Phase  Figure 5.7 shows the training time and the number of trials the participants required to learn the four tactile alert schemes at each stage. We observed in all cases an increase in the training time as the number of tactile alerts increased. The tactile alert scheme S1[ ,o] appeared to require additional training time. 2 Since the time to convey the tactile alerts was different between schemes, we used the number of trials attempted to learn the tactile alert in each stage for the analysis. A five-way ANOVA revealed a statistical significance in the number of trials attempted at each stage of training [F 71  =  11.85, p < 0.001] and the tactile  (j=  -----S1 32  30  =0) 0) ‘‘‘S3 3 cr 1.41) “‘S45. rl.52)  ftr. = 3.  /  20 16 10  25  3  rr rate (bit)  3.5  4  4,5  Figure 5.7: Mean training time (left) and mean number of trials required (right) to learn the tactile alert scheme at each stage. The four tactile alert schemes are represented by the encoding parameter [ii, u].  alert scheme used [F  =  21.42, p < 0.001]. As expected, a post-hoc Tukey’s HSD test  showed that the participants required more trials to learn how to differentiate the tactile alerts as the number of tactile alerts increased. Participants required more ,o} (mean ± SD: 17.43 ± 0.88), 2 trials to learn the alerts in tactile alert scheme S1[ ,o] (10.92 ± 0.78), S3[ 3 411 (11.36 ± . 1 , 3 as compared to the other three schemes [S2[ 0.67), and S4[s,i. ] (8.82 ± 0.68)]. Participants also required more trials to learn 52 the alerts in ] 41 than in ] . 1 , 3 S3[ 52 (Figure 5.8). . 1 , 5 S4[ Significant differences were observed in gender [F age group of the participants [F  =  =  18.55, p < 0.001] and  30.28, p < 0.001]. A post-hoc Tukey’s HSD test  further demonstrated that the female participants (13.67 ± 0.57) required more trials to learn the tactile alerts than the male participants (10.60 ± 0.48). Participants above the median age of 24 yr (14.18 ± 0.55) required more trials to learn the tactile alerts than those who were below or at the median age (10.08 ± 0.53). The analysis also demonstrated a trend in the effect of musical background of the participants [F =  3.49, p  =  0.0622] on the number of trials required to learn the tactile alerts.  72  .  15  I  z  10  Tactile Alert Scheme  Figure 5.8: Post-Hoc Thkey’s HSD test on the tactile alert scheme for training. The number of trials attempted to learn the tactile alert in each stage was used in the analysis. Statistical difference is illustrated by the star (*) symbol. 5.4.2  Testing Phase  Accuracy The accuracy of tactile alert identification, measured at each stage, for the four tac tile alert schemes is shown in Figure 5.9. The accuracy of tactile alert identification for Si [2,0] appeared to decline more rapidly than the other three tactile alert schemes as the number of tactile alerts increased. Alert scheme  [3,141]  demonstrated the  highest accuracy, compared to all tactile alert schemes tested, at all stages of the study. A five-way ANOVA illustrated a statistical difference in the accuracy for each stage of the study [F  =  28.52, p  between tactile alert schemes [F  <  0.001]. A statistical difference was also observed 117.3, p  <  0.001]. A post-hoc Tukey’s HSD test  showed that the accuracy deteriorated as more tactile alerts were conveyed. The accuracy of identifying the tactile alerts in Si[20] (mean + SD: 67.85% + 1.18%) was lower than that in other three schemes [52[3,o[ (78.77% ± 1.05%), 53[3,1.41[ (94.02% ± 0.91%), and  S[5152[  (89.50% + 0.91%)] (Figure 5.10).  73  The five-way ANOVA revealed a statistical difference in the accuracy be tween gender [F  =  7.48, p  =  0.00641. The female participants demonstrated a lower  accuracy in tactile alerts identification (81.23% ± 0.77%) than the male participants (83.84% ± 0.65%). No statistical difference was detected in music background [F =  2.63, p  =  0.106] and age group [F  =  0.64, p  =  0.424] on accuracy of tactile alert  identification.  z:;S3(3;o a 70  -  -  7  .:  I..  I0___  10  —‘-—S1 (=2;a0) S2(p.3;a0) 53(3a141) S4(5;a1.52  2  l 15  2  25 3 IT rate (bit)  141) 1 52)  4  35  15  2  3 2.5 IT rate (brt)  3.S  4  4.5  Figure 5.9: Mean accuracy (left) and response time (right) for tactile alert identifi cation with the four tactile alert schemes. A five-way ANOVA demonstrated a lower as compared to the other three accuracy of identifying the tactile alert in Si schemes (S2[3,o], ], 41 and 52 . 1 , 3 S3[ ,i The vertical dashed-line indicates the IT 5 S4[ ]). rate at 2.8 bits.  Response Time Figure 5.9 shows the response time of tactile alert identification, at each stage, for the four tactile alert schemes.  Not surprisingly, a longer response time was  observed when the number of tactile alerts conveyed in the testing phase increased. Participants required the longest time to identify the tactile alerts in S4[ ]. We 5152 observed the shortest response time when S3[ ]was used. 3141 A five-way ANOVA detected a statistical difference in response time at dif ferent stages of the study [F schemes [F  =  =  27.97, p > 0.001] and between the tactile alert  153.93, p > 0.001]. The post-hoc Tukey’s HSD test showed that the 74  140 1212  60 412 20  Tactile AleS Scheee  Tactile  Alert  Scheme  Figure 5.10: Results of post-hoc Tukey’s HSD test on accuracy (left) and response time (right) of tactile alert identification for the four tactile alert schemes. Statistical difference is shown by the star (*) symbol. participants required less time to identify the alerts in S1[ ,o] (mean ± SD: 4.32 2 ± 0.09s) and 4 ] (5.22 ± 0.08s) and 30 ,i. 3 S3[ i ] (4.27 ± 0.07s) when compared to S2[ ,i. (6.11 ± 0.08s). Participants also demonstrated a faster response time when 5 S4[ ] 52 identifying alerts in S2[ ]than in 52 30 ,i. (Figure 5.10). 5 S4[ ] We detected a statistical difference between gender [F  =  7.81, p  =  0.0051  on the response time of tactile alert identification. The male participants (4.88 ± 0.05s) were found to respond faster than the female participants (5.08 ± 0.06s). No significant effect on response time was found in music background [F 0.675] and age group [F  =  0.12, p  =  =  0.18, p  =  0.733].  Theoretical Optimal IT Rate The maximum likelihood estimation IT (ITest) , which represents the error-free trans mission of tactile alerts, was determined from the confusion matrix constructed in the final stage of the study with IT rate  =  4.32 bits (20 tokens). Table 5.4.2 sum  marizes ITest (in bits) and the corresponding tactile alerts (in token) for the four tactile alert schemes described. The highest ITest of 3.73 bits (13.26 tokens) was ,o]. 2 found in ] 41 and the lowest IT . 1 , 3 S3[ 08 of 2.07 bits (4.21 tokens) in S1[ 75  Scheme ,o] 2 Si] S2[3,o] ,i. 3 S3[ ] 41 84[5,1.52]  ITest (bit)  2.07 2.65 3.73 3.25  No. of Token 4.21 6.30 13.26 9.48  Table 5.1: Maximum likelihood estimation IT rates (ITest) for error-free tactile alert transmission for the four tactile alert schemes. IT t is expressed in bits and the 63 corresponding tactile alerts in tokens. ITest was determined by the confusion matrix constructed in the final stage of the study at IT rate = 4.32 bits (20 tokens). Tactile Perception around ITthreshold = 2.8 bits  Figure 5.11 shows the accuracy below the ITthreshjd  =  2.8 bits. The accuracy of  tactile alerts identification, regardless the choice of tactile alert scheme, was found below the hypothesized accuracy of 99%. Participants attained an accuracy of 90% or above in identifying the alerts in 41 ] 20 ,i. and S4[5,1.52] but the accuracy in Si[ 3 S3[ ] was relatively poor. The accuracy of tactile alert identification when the IT rate was above the ITthreshold is illustrated in Figure 5.12. Si [20] had the highest decay rate in accuracy  (-21.56%  /  bit) while S3[3,1.41] had the lowest rate of decay (-4.32%  /  bit). Results  of the linear regression suggest that the accuracy decayed linearly with the IT rate in S1[ ,o] and 52[3,oJ, but not in S3[3,1.41] and 2  5.5  S4[5,1.52].  Discussion  This study explored an approach to the design of an effective rhythm-based tactile alert scheme applied on the abdomen. We have tested the amount of information that can be encoded in different tactile alert scheme designs without sacrificing the accuracy of tactile alert identification. We classified the alert schemes using the encoding parameters  [j,  °i  and found that S3[3,1.41] gave the highest accuracy and  fastest response time in tactile alert identification. The training time was shortest for  S[5,1.52]  followed by 52[3,o],  S3[3,1.41]  and S1[2,o].  76  100  80  —G--—31(20) S2 ( = 3; = 0) S3(=3;c=1.41) S4(I.=5; c= 1.52)  75 1  1.2  1.4  1.6  1.8 2 IT rate (bit)  2.2  2.4  2.6  2.8  Figure 5.11: The accuracy of tactile alert identification for the four tactile alert schemes below ITthreshold = 2.8 bits. The two dashed lines represent the accuracy of tactile alert identification in the range between 99% and 100%. We found that the choice of encoding parameters was likely to impact the training time, accuracy and response time. We observed a short training time when the value of  j.t  was small (e.g.  rapidly after ITthreshold  =  =  2), but the accuracy appeared to deteriorate  2.8 bits. The use of a > 0 to smooth the rate of decay in  accuracy after ITthreshold was found. Below the ITtkreshold, 41 ,i. and S4[s,i. 3 S3[ ] ] 52 demonstrated a satisfactory accuracy of above 90% in tactile alert identification, even though it was lower than the hypothesized accuracy of 99%. In terms of the maximum likelihood transmission rate, 411 ,i. demon 3 S3[ strated the highest ITest of 3.73 bits. This showed that, theoretically, 13.26 tactile alerts could be conveyed and identified by the tactile belt prototype users without error. ITest was found to increase when a> 0 and would be highest when 0  [/2  =  3,  > 0].  The post-hoc analysis demonstrated a statistical difference between gender and age group in both accuracy and response time. Male participants demonstrated 77  100  -%  SIope-973 %/bft  -..,  To  N  50  (j2; a0)  SIope-21.66%ibrt r0.975  —52(it=3;a=0) 4 — ‘S3(jt3;141) “S4(t5; a1.52) 4[  2.6  I  I  I  I  I  I  I  2.8  3  3.2  3.4  3.6 0’ rate (bit)  3,8  4  I  4.2  4.4  4.6  Figure 5.12: The accuracy of tactile alert identification for the four tactile alert schemes above ITthreshojd = 2.8 bits. The slope and the R 2 value for each accu racy line are summarized in the legend. The linear approximation of the curve is considered valid if R 2 > 0.975.  78  a higher accuracy and faster response time in identifying the tactile alerts than female participants. A similar result was also found in the age group, in which par ticipants under the median age showed a better performance in tactile alert iden tification. Interestingly, formal music education did not appear to have significant effect on accuracy and response time. In summary, this study provided guidelines for designing an effective tactile alert scheme to facilitate a high IT rate. We concluded that encoding parameters of  [t  =  3, o  >  0] should be used in the design of the tactile alert scheme, while  encoding parameters of [> 3 , u  =  0] should be avoided. Implications and findings  in this study would provide clearer direction for the design of an effective tactile alert scheme to encapsulate complex physiological information.  79  Chapter 6  User Study #4: Performance of Tactile Belt Display under Simulated Low and High Clinical Workload Conditions 6.1  Introduction  The three studies presented in Chapter 3 to Chapter 5 focused on exploring the type of tactile display prototypes and the design of the tactile alert schemes that would be suitable for our clinical application. We found that participants preferred mechanical vibration over electrical stimulation as the medium to convey tactile alerts (Chapter 3).  We also showed the use of Tacton parameters, particularly  the rhythm and spatial location on the abdomen, to design a complex tactile alert scheme that could encode all necessary physiological information (Chapter 4). We further demonstrated that rhythm-based tactile alert scheme with encoding param eter of  [ji =  3, o > 0] would preserve high accuracy and a rapid response time in  80  tactile alert identification (Chapter 5). In this chapter, we further our investigation by inviting certified anesthesi ologists and anesthesia residents to assess the usability of a tactile belt prototype under low workload (LW) and high workload (HW) simulated clinical monitoring conditions. We evaluated how well (i.e. the accuracy) and how fast (i.e. the re sponse time) the anesthesiologists could identify the tactile alerts while performing different levels of physiological monitoring tasks.  6.2 6.2.1  Methods Tactile Display Prototype  The tactile belt prototype (Figure 6.1) was developed by British Columbia Institute of Technology (BCIT; Burnaby, B.C., Canada) and was powered by a rechargeable battery compartment. The prototype was worn on the abdomen of the user and consisted of six evenly spaced tactors (DC motors). The tactors operated at 100 Hz, which provided a large amplitude of vibration that could be sensed by the human skin. Vibrotactile alerts were conveyed through the belt prototype using BluetoothTM technology.  6.2.2  Tactile Display Location  Findings in Chapter 4 suggested that, in theory, a 100% accuracy in tactile alert identification could be preserved when four spatial locations on the abdomen were used to receive tactile alerts. In this study, therefore, we used four out of the six tactors to convey tactile alerts, forming an anterior and a posterior cluster, with two tactors per cluster (one on each side). Each tactor was used to represent a distinct physiological parameter, namely exhaled end-tidal carbon dioxide tension ), peak airway pressure (Ppeak), respiratory minute volume (MVexp), and 2 (EtCO mean noninvasive blood pressure (NIBPmean). Using the umbilicus as a point of  81  reference at the 12 o’clock position, Ppeak was represented by the tactor located 2 at the 4 o’clock position, MVexp at the 8 o’clock at the 2 o’clock position, EtCO position, and NIBPmean at the 10 o’clock position. The tactors located at the 12 o’clock and 6 o’clock positions were not used (Figure 6.1). Navel Blood Pressure  Airway Pressure  I TactO,  End Tidal Carbon Dioxide  Minute Volume  Spine  Figure 6.1: Tactile belt display prototype (left) and representation of physiological parameters on the prototype (right). In this study, four out of the six tactors were used to convey vibrotactile alerts. Each of the four tactors represented one physiological parameter: EtCO , pPeak, MVexp, and NIBPmean. The tactile alerts 2 were conveyed to the tactile belt prototype user using BluetoothTM technology.  6.2.3  Tactile Alert Scheme  The tactile alert scheme consisted of four tactile alerts that were used to represent the changes in the four physiological parameters.  The alerts were described by  three components: physiological parameter (EtCO , Ppeak, MVexp, or NIBPmean), 2 direction of change (increasing or decreasing), and level (level 1 or 2). The direction of change and the level were represented by four distinct combinations of short (200 ms) and long (1200 ms) vibrating pulses (rhythm) (Figure 6.2). The change in a particular parameter was described by a tactile alert conveyed through the corresponding tactor location. For example, an “increasing, level 1” alert in EtCO 2 was represented by conveying an “increasing, level 1” alert through the tactor at the 2 o’clock position (Figure 6.1). The alerts were generated at random time and in random order by the software developed for this study.  82  l200ms  200ms  Increase level 1  Increase level 2  j1J[ JIJ1JIJIJ[’: ,  Decrease level 1  Decrease level 2  Figure 6.2: Tactile alert scheme. The scheme consisted of four distinct tactile alerts used to represent a change in the four physiological parameters. The alerts were generated by varying short (200 ms) and long (1200 ms) vibrating pulses. The changes were described by the direction of change (increasing or decreasing) and level (level 1 and level 2).  6.2.4  Think Aloud Method  Participants were instructed to identify the tactile alerts using the think aloud method in this study (Section 6.2.7). The think aloud method involves participants vocalizing their thoughts as they are performing a set of specified tasks [26]. This method has been used extensively in the evaluation and design of user interfaces. We has chosen the think aloud method over other usability assessment methods be cause it represents the least obtrusive method of assessing the usability of the belt prototype.  6.2.5  Workload  Human workload can be defined as the ratio of resources required to perform a set of tasks to the resources available to the individual at a given time [33]. If the ratio approaches or is beyond 100%, the individual is said to be overloaded; on the other hand, if the ratio approaches 0%, the individual is regarded as underloaded. Workload assessment tool such as the NASA Task Load Index (TLX) [45] was used to evaluate the amount of workload. This tool is useful in assessing the mental workload of the participants when they were performing specific tasks. In this study, the low and high workload conditions were defined based on the  83  number of physiological parameters participants were required to monitor (i.e. track the changes) on a simulated anesthesia visual display. The low workload condition (LW) was defined as tracking only the changes in the heart rate on the visual display, whereas the high workload condition (HW) involved tracking the changes in heart rate, oxygen saturation, respiratory rate, and body temperature. In addition to this tracking task, the participants were required to identify the tactile alerts using the think aloud method. A pilot study which recruited 12 non-medically trained participants was con ducted to validate the difference in workload between the two workload conditions (LW and HW). Using the NASA TLX workload assessment tool, we found that the workload was 15.97 and 38.56 for LW and HW conditions, respectively. A Krustal Wallis ANOVA demonstrated a statistical difference between the two conditions  2 [x  =  6.2.6  lO.135,p  =  0.0011].  Confusion Matrix  We constructed two sets of confusion matrices for the physiological parameter (spa tial location) and the tactile alerts (rhythm), one set for each workload condition. The modified confusion matrix described in Chapter 4 was used. 6.2.7  Study Procedures  The study was approved by the Clinical Research Ethics Board of The Univer sity of British Columbia (Board Approval Number: H07-02044) and The Children’s & Women’s Research Review Committee (Board Approval Number: CWO8-0050). Participants in the study were either certified specialist anesthesiologists or anes thesia residents from hospitals in Vancouver. The study was approximately 60 mm in duration and was divided into a training phase, a testing phase, and a post-study evaluation phase (Figure 6.3). Participants were enrolled to test the belt prototype on their abdomen under two workload conditions (LW and HW). Each participant  84  received an honorarium of CAD$5.00 coffee shop gift card.  Figure 6.3: Flow diagram of the study procedures. The duration of the study was approximately 60 mm and was divided into a training phase, a testing phase, and a post-study evaluation phase. Participants in this study were either staff anesthesiologists or anesthesia residents. They were instructed to test the tactile belt prototype under LW and HW conditions. Following written consent, participants took part in a two-phase training session. The first phase familiarized the participants with the tactile alert scheme and the think-aloud method. We facilitated an effective training by providing the participants with a detailed explanation on the design of the alert scheme prior to the training session. The participants then completed a post-training quiz, which tested their ability to identify the tactile alerts using the think-aloud method. They were required to identify the alert by indicating three components: (1) the type of physiological parameter the alert represented, (2) the direction of change, and (3) the level of the alert. A total of 480 alerts were conveyed in the quiz. This training phase was terminated when participants obtained an accuracy of at least 75% in the quiz. Participants who failed to achieve the required accuracy after a training period of 30 mm  were no longer eligible to participate in the study. We measured  the duration of training as the amount of time the participants required to learn the tactile alert scheme in order to obtain at least 75% in the post-training quiz. The number of attempts the participants required to pass the post-training quiz was also  85  recorded. A second training phase allowed participants to become familiar with the graphical user interface (GUI) of the simulated anesthesia monitor (Figure 6.4). During the testing phase, all participants performed two tests. Each test corresponded to one workload condition (LW or HW) and was 10 mm in duration. The tests were arranged in random order to reduce bias using a random permutation function randperm in Matlab (Mathworks Inc., Natick, MA, USA). In each workload test, the participants were instructed to perform two tasks: (1) the tracking task and (2) the tactile alert identification task. Participants tracked the changes of the physiological parameters on the simulated anesthesia visual display by clicking the appropriate button on the GUI in the tracking task, and identified any tactile alerts conveyed through the belt prototype using the think aloud method in the tactile alert identification task. The number of physiological parameters required to be monitor depended on which test the participants were performing (Section 6.2.5). We used an audio recorder (ICD-B500, Sony, Japan) with an external microphone (CM-Ph, AIWA, Japan) to record the responses of the participants in both the post-training quiz and the testing phase. A total of 958 and 956 tactile alerts were conveyed in the LW and HW tests, respectively. The accuracy and response time in tactile alert identification were measured in the testing phase. The accuracy was defined as the ratio of number of correctly identified alerts to the total amount of alerts transmitted. The response time was defined as the latency between the activation of the tactile alert and the time at which the participants identified any one of the three components of the tactile alert in a complete response (i.e. all three components were identified). The number of alerts the participants failed to identify (i.e. missed alerts) under each workload condition was also recorded. Alerts were classified as missed if the participants detected the presence of the alerts but failed to identify the three components of the alert, or if the participants failed to identify the alert before the next alert was sent to the belt prototype. Missed alerts were treated as incorrect responses with a  86  lb...  Ofl.mflDI  IflOIUGf  l2tll4lllC  I  ii ..,“,  1 no:I  J  III  I  I  -1  I!’  !IJ  Figure 6.4: Screen shots of the tracking task in LW (left) and HW (right) condition. Participants were asked to perform a tracking task by monitoring the heart rate in LW condition. They were required to monitor heart rate, oxygen saturation, respiratory rate, and body temperature in HW condition. response time equal to the elapsed time between the activation of the current alert and that of the next alert. The accuracy, response time, and missed response for the tracking task were measured in both the LW and the HW conditions. We defined the accuracy as the number of changes recognized correctly by the participants, and the response time as the time between the change in the parameter and the instant the participants clicked the appropriate button on the GUI. The missed response referred to the situation where the participants failed to notice a change in the simulated anesthesia visual display. We administered a Computer Usability Satisfaction Questionnaire (CUSQ) [74] to measure the satisfaction of the participants with the belt prototype in the post-study evaluation phase. The questionnaire consisted of 19 statements. Par ticipants could express their level of agreement with each statement by ranking on a scale of 1 to 7, with 1 indicating “strongly disagree” and 7 meaning “strongly agree.” If participants chose not to respond to a particular statement, they would  87  check the “not applicable” (N/A) box. Responses marked as “N/A” were removed from the analysis.  6.2.8  Statistical Analysis  The sample size calculation for this study was based on the previous results in accu racy of tactile alert identification [87]. We conducted apriori power analysis targeted to detect a 5% difference in the accuracy between the two workload conditions at cv  =  0.05 and  =  0.2. A sample size of 30 participants would be required. The  analysis of variance (ANOVA) was used to test for differences between the workload condition (LW or HW) with respect to the accuracy and response time of tactile alert identification. Differences were considered significant at p < 0.05. The choice of whether to use non-parametric or parametric statistical test followed the crite ria presented in Section 3.3. Outliers identified using the 1.5 inter-quartile-range (IQR) technique [108] were removed from the analysis only if the data distribu tion was normal. This was to prevent outliers from skewing the results in a set of normally distributed data [84]. Yet, for non-normally distributed data, we did not remove possible outliers to avoid discarding representative data from a long-tailed, non-normal distribution. The mean responses for each statement in the post-study questionnaire were calculated by averaging the responses that were not marked as  “N/A”. The analysis was performed in Matlab (Mathworks Inc., Natick, MA, USA).  6.3 6.3.1  Results Study Participants  We recruited 30 staff anesthesiologists and anesthesia residents (22 male and 8 fe male) in this study, with 24 participants (18 staff, 6 residents) from the British Columbia Children’s Hospital and 6 participants (5 staff, 1 resident) from St. Paul’s Hospital. The age of the participants ranged from 28 to 63 yr, with 37%, 30%, and  88  33% of participants in 28-35 yr, 36-45 yr, and above 46 yr age group, respectively. All participants completed the tests under both LW and HW conditions.  6.3.2  Training phase  A total of 30 measurements (30 participants x 1 training) were recorded but only 29 measurements were analyzed, since one set of data was accidentally erased. Four possible outliers were detected but remained in the analysis, since the data were non-normally distributed (Shapiro-Wilk test, p < 0.001).  12 10  0  2 0  126  76 225 Tmining Time (s)  275  326  Figure 6.5: Bar chart demonstrating the distribution of the required training time for the 29 participants. Four possible outliers, identified using the 1.5 IQR technique and highlighted by the squared box on the bar chart, were not removed from the analysis since the data distribution was non-normal. The mean (± SD) training  time was 123.8 ± 71.ls. The participants required a mean (± standard deviation (SD)) training time of 123.8 ± 71.ls to familiarize themselves with the tactile alert scheme (Figure 6.5). No outliers were detected in the accuracy and response time to tactile alert identification in the post-training quiz. Participants attained a mean accuracy of 92.9% ± 8.3% and missed 0.83% of the total transmitted tactile alerts (i.e. 4 alerts) in the quiz. A mean response time of 10.7s ± 3.2s was required to identify the tactile alerts. 89  Accuracy (%) Missed alerts Response Time (s)  Low Workload 91.9 ± 10.8 0.77 ± 1.01 9.4 ± 0.9  High Workload 88.8 + 12.2 1.20 ± 1.67 9.7 ± 0.9  p-value 0.128 0.245 0.095  Table 6.1: Summary of ANOVA results in accuracy, missed alerts, and response time in the testing phase. Data are expressed as mean ± SD where appropriate. All but two participants completed the first phase of training in one attempt. Both participants who failed on the first attempt confused the mapping of the phys iological parameters with the corresponding spatial locations on the belt prototype. Both participants passed the quiz at the second attempt. Testing phase  6.3.3  Accuracy We analyzed 60 measurements (30 participants x 2 workload tests) for accuracy. The data were non-normally distributed (Shapiro-Wilk test, p and p  =  =  0.001 for LW test  0.003 for HW test) and did not appear to follow any of the well-known  parametric distributions.  Five possible outliers were identified (Figure 6.6) but  remained in the analysis, since the data were non-normally distributed, An accuracy of 91.9% ± 10.8% was observed for the LW test and 88.8% ± 12.2% for the HW test. A Kruskal-Wallis ANOVA detected no statistical differences  2 [x  =  2.31, p  =  0.128]  in the accuracy of tactile alert identification between the two workload conditions (Table 6.1). Missed alerts Participants have missed 23 tactile alerts (2.4% of the total transmitted alerts) in the LW test and missed 36 tactile alerts (3.8% of the total transmitted alerts) in the HW test. The data were non-normally distributed (Shapiro-Wilk test, p  90  <  0.001 for both  LW and HW tests) and did not appear to follow any of the well-known parametric distributions. A Kruskal-Wallis ANOVA detected no statistical difference in the number of missed alerts between the two workload conditions  2 [x  =  1.35, p  =  0.245]. Response time A total of 60 measurements (30 participants x 2 tests) were collected for the response time. Three possible outliers were identified but retained in the analysis (Figure 6.6), since the data distributions were non-normal (Shapiro-Wilk test, p  =  0.003 for  both LW and HW tests). The data did not appear to follow any of the well-known parametric distributions. The mean response time to identify the tactile alerts was 9.4 ± 0.9s in the LW condition and 9.7 ± 0.9s in the HW condition. No statistical difference was detected between the two conditions (Kruskal-Wallis ANOVA,  2.79, p  =  2 [x  =  0.095]).  H 86 —  .95  —  E  LW  11W W11ktOod Condihon  Wo,kload Condition  Figure 6.6: Accuracy (left) and response time (right) of tactile alert identification under low and high workload conditions. The box-whisker plots depict the upper quartile, median, and lower quartile values (top, middle, and bottom lines of the box), respectively. Possible outliers were identified using the 1.5 IQR technique and marked as “+“. LW = low workload; HW = high workload. Outliers were remained in the analysis since the data were non-normally distributed  91  Confusion Matrix The modified confusion matrix for the physiological parameters (spatial location) and the tactile alerts (rhythm) are shown in Figure 6.7. The missed tactile alerts were excluded from the construction of the confusion matrix. #4  NivM  #2  55  C.5  I  I 1 EE  54  5551 LW ?cIrtwi&f3r  H’V ?rMI12rS  LW Affrl. ScIie  H’AeIt.SehM  Spatial Location  #1  IncrI.vI1  #2  Incras.I.v.I2  #3  I I  #4  I2  flJIj[  Rhythm  Figure 6.7: Modified confusion matrix for physiological parameters (spatial location; top) and tactile alerts (rhythm; bottom) for LW and HW conditions. The numbers in the modified confusion matrices correspond to the number on the reference diagrams of spatial location and tactile alert scheme (left). Participants appeared to encounter most difficulties in distinguishing the tac 2 location tile alerts transmitted to the Ppeak location from those sent to the EtCO under both LW and HW conditions. Participants also demonstrated the highest con fusion rate by misinterpreting the Increasing Level 2 alert as the Increasing Level 1 alert in the LW condition, and Decreasing Level 3 alert as Increasing Level 2 alert in the HW condition.  92  Measurement Response Time (s) Accuracy (%) Alert missed (%)  HR 4.8 ± 1.8 100 0  Table 6.2: Performance of participants in the tracking task under LW condition. Participants were required to monitor the change in heart rate (HR) on the simulated anesthesia visual display. The data are expressed as mean ± SD where appropriate.  rMeasurement Response Time (s) Accuracy (%) Alert Missed (%)  HR 9.1 + 4.3 100 ± 0 0.5  Sp0 2 5.8 ± 2.8 91.4 ± 11.6  7.6  RR 11.5 ± 4.7 100 ± 0 3.3  T 11.5 ± 6.4 100 ± 0 5.0  Table 6.3: Performance of participants in the tracking task under HW condition. Participants were asked to track the changes in heart rate (HR), oxygen satura tion (Sp0 ), respiratory rate (RR), and body temperature (T) on the simulated 2 anesthesia visual display. The data are expressed as mean ± SD where appropriate.  Tracking Task Table 6.2 and 6.3 summarize the performance of the participants in the tracking tasks under the LW and HW conditions, respectively.  6.3.4  Computer Usability Satisfaction Questionnaire (CUSQ)  Table 6.4 summarizes the responses collected from the questionnaire.  6.4  Discussion  We observed that the medically trained participants can quickly become familiar with the tactile alert scheme. The anesthesiologists’ high degree of accuracy and rapid response time in identifying tactile alerts demonstrated the feasibility of using the tactile belt prototype as a supplemental advisory device under different clinical  93  Statement Overall, I am satisfied with how easy it is to use this system. It is simple to use this system. I can effectively complete my work using this system. I am able to complete my work quickly using this system. I am able to efficiently complete my work using this system. I am able to efficiently complete my work using this system. It was easy to learn to use this system. I believe I became productive quickly using this system. The system gives error messages that clearly tell me how to fix problems. Whenever I make a mistake using the system, I recover easily and quickly. It is easy to find the information I need. The information provided with the system is easy to understand. The information is effective in helping me complete my work. The interface of this system is pleasant. I like using the interface of this system. This system has all the functions and capabilities I expect it to have. Overall, I am satisfied with this system.  Response 5.60 (1.35) 5.50 (1.41) 4.55 (1.80) 4.93 (1.65) 4.93 (1.70) 4.83 (1.60) 5.63 (1.19) 5.25 (1.43) 2.90 (2.27) 3.38 (1.82) 5.11 5.96 4.77 5.21 4.77 4.38  (1.71) (0.82) (1.63) (1.55) (1.41) (1.77)  4.67 (1.56)  Table 6.4: Summarized Responses from CUSQ. The responses were ranked on a nominal scale of 1 to 7, with 1 indicating “strongly disagree” and 7 “strongly agree”. The participants could check the “not applicable” (N/A) box if they chose not to respond to a particular statement. “N/A” responses were removed from the analysis. Data are presented as mean (SD) as appropriate.  94  workload conditions. Training time can be brief if participants are provided with a detailed ex planation of the tactile alert scheme design prior to the actual training. We found that the amount of missed alerts, accuracy, and response time in the tracking task deteriorated as the workload increased. In contrast, we detected no statistical dif ferences in the accuracy, response time, and amount of missed tactile alerts between the LW and HW condition when the participants used the tactile belt prototype. The response time to identify tactile alerts (LW  =  9.4s; HW  =  9.7s) was found  moderately slower than detecting the audio alerts (3s), but much faster to identify alerts displayed on the visual monitor (40s) [83]. This confirms the potential value of using tactile technology to improve the communication to the anesthesiologists on adverse physiological changes occurred in the patients, especially in high workload conditions. The close-to-significant difference detected in the response time of tactile alert identification (p  =  0.085) may suggest that the information transmitted by the  belt prototype could potentially lead to information overload in some individuals. From a usability standpoint, participants were satisfied with the interface of the belt prototype and expressed that the tactile alert scheme was easy to learn. We learned from the informal, post-study discussion that some participants suggested reducing the amount of information conveyed by the belt prototype. Some also suggested incorporating more important physiological parameters, such as heart rate or respiratory rate, in the tactile alert scheme. We also observed an interesting trend in the anesthesiologists that their adoption of the tactile display declines with years of anesthesia practice. Senior anesthesiologists tended to favor the use of standard clinical monitoring and appeared to be more reluctant to adopt new monitoring technologies, as compared to anesthesia residents. Nevertheless, participants agreed that our tactile belt prototype could facilitate better communication of abnormal changes in patient’s physiological status in the operating room.  95  Chapter 7  Conclusion and Future Work We explored the potential use of tactile technology to convey physiological infor mation of anesthetized patients to the attending anesthesiologists in the operating room. Four laboratory studies were described to evaluate different aspects of the tactile display prototype and the corresponding tactile alert scheme. The method of tactile stimulation, the approach to design a complex tactile alert scheme, and the performance of a tactile belt display in transmitting tactile alerts to users un der different simulated clinical workload conditions were investigated. The major findings and implications derived were as follows: In the first study, the participants demonstrated a preference for using me chanical stimulation to convey tactile alerts. Vibrotactile interface was more pleas ant than the electrotactile interface and the participants demonstrated a higher accuracy in identifying the vibrotactile alerts. No statistical difference was detected in the accuracy and the response time of tactile alert identification when the vibro tactile alerts were conveyed either on the forearm or the wrist. The second study evaluated a tactile alert scheme with 36 distinct tactile alerts designed using Tactons [7j. Three types of rhythm, two types of roughness, and six spatial locations were used to construct the tactile alert scheme. We de termined from the modified confusion matrix that 18 distinct tactile alerts can be  96  conveyed and identified correctly by the participants with this tactile alert scheme design. Further analysis demonstrated that we can improve the accuracy of tactile alert identification by using only the rhythm and the spatial location. The third study explored the design of a rhythm-based tactile alert scheme for encoding complex tactile information. Four distinct tactile alert schemes, classi fied using the encoding parameters  [pt,  o], were evaluated and compared in terms of  training time, accuracy, and response time of tactile alert identification. The high est tactile information transmission rate was detected in the tactile alert scheme designed with encoding parameters  [=  3, cr > 0].  The fourth and final study assessed the ability of the tactile display users in identifying the tactile alerts under low and high simulated clinical workload condi tions. Findings from previous studies were integrated in the experimental design. The participants demonstrated a lower accuracy and slower response time in the tracking task in the high workload condition, but they did not encounter difficulty in the tactile alerts identification task with increased workload. This confirms the potential value of using the tactile display to improve the communication to the anesthesiologists on adverse physiological changes occurred in the patients, espe cially during high clinical workload.  7.1  Limitations  The four studies described in this thesis were pre-clinical and performed in a lab oratory setting. The test environment was under controlled and the testing time was short (less than one hour). Results demonstrated the potential use of tactile display to improve the communication to the anesthesiologists on adverse physio logical changes occurred in the patients. Further assessment with a much longer duration of testing is needed to validate the use of the tactile display in the clinical environment. We have not identified the most suitable tactile stimulation location for our 97  application. Three tactile stimulation locations, namely, the wrist, the forearm, and the abdomen, were tested in this thesis. The choice of the tactile stimulation locations was based on previous studies.  Except for evaluating the vibrotactile  localization on the wrist and the forearm, no simultaneous comparisons were carried out to evaluate the tactile perception between the three stimulation locations.  7.2  Future Work  Possible directions for future work include (1) evaluating the tactile display in the clinical environment, (2) assessing the tactile alert scheme in the clinical environ ment, (3) exploring other tactile stimulation locations, and (4) adopting the tactile display for clinical use.  7.2.1  Evaluation of Tactile Display in the Clinical Environment  The tactile display can be evaluated in a less controlled environment with a much longer testing time, e.g. weeks, in the clinical environment. The tactile display is intended to serve as a supplemental advisory device for physiological monitoring and therefore it should be possible for the anesthesiologists to use it in all surgical cases. Long hours of testing in the clinical study can facilitate assessments in problems such as durability of the tactile display, sensation adaptation to tactile alerts, and users’ preferences in terms of comfort and level of frustration, etc. More meaningful tactile alerts using real clinical data can be conveyed to the study participants during the clinical study. We will also evaluate the use of the tactile display alongside standard monitoring. Factors such as the training time, accuracy and response time of tactile alert identification will be collected and compared between the situation when the tactile display and standard monitoring are used, and the situation when only the standard monitoring is practiced. This will identify any improvement in the communication to the anesthesiologists of adverse physiological changes in the patients. 98  Critical situations when many physiological changes are likely to occur will be simulated. Such situation will trigger frequent transmission of tactile information from the tactile display to the users. We will assess the level of annoyance, frustra tion, and distraction that may be caused by the tactile display. The possible use of the tactile display in other clinical units such as the intensive care unit should also be investigated. Our goal is to develop a tactile display that can improve the com munication to the anesthesiologists on adverse physiological changes in the patients, while minimizing additional frustration and distraction.  7.2.2  Evaluation of Tactile Alert Scheme in the Clinical Environ ment  Some anesthesiologist participants indicated that too much information were en coded in the tactile alert scheme (Chapter 6). They expressed that the problem of information overload would increase their level of frustration since identifying the tactile alerts was not their only task. Results presented in Chapter 5 appear to con tradict this opinion. Analysis of the confusion matrix indicated that an information transmission rate of 3.73 bits (13.26 tokens) could be transmitted and identified correctly by the tactile display users if the tactile alert scheme was designed using the encoding parameters  [  =  3, u  >  0]. The tactile alert scheme in Chapter 6  consisted of only four distinct tactile alerts (i.e. 4 tokens) and therefore should not cause information overload to the tactile display users. Such contradiction between the opinion of the participants and our findings may be due to the short testing time of the tactile alert scheme in the laboratory study. Participants spent, on average, 123.8 s to learn how to use the tactile display, and they spent, approximately, an other 20 mm  in using the tactile display in the testing phase. The anesthesiologist  participants are unlikely to adapt to this new physiological monitoring device within such a short period of time. We should delay the conclusion of information overload until after the participants have tested the tactile display and the corresponding  99  tactile alert scheme for a longer period of time (e.g. few weeks). Some tactile alerts maybe conveyed less frequently than other alerts in certain surgical cases. Tactile display users are likely to forget the meaning encapsulated by these tactile alerts after a prolonged period of time. Future investigations should evaluate the ability of participants in identifying tactile alerts that are transmitted less frequently. This will reveal the need of regular tactile alert scheme re-training to refresh the memory of the tactile display users the meaning of each tactile alert. Anesthesiologist participants indicated that the tactile alerts should convey more important physiological parameters such as heart rate and oxygen saturation 2) (Chapter 6). We intentionally avoided conveying these physiological pa 0 (Sp rameters through the tactile display because these parameters are well presented using the sonification pulse oximeter [117]. Conveying the already well presented physiological information appears to contradict with our intended use of the tactile display as a supplemental advisory device. We will consult certified anesthesiol ogist specialists to identify additional physiological parameters suitable for tactile communication. These parameters will be appended to our existing tactile alert scheme.  7.2.3  Exploration of Other Tactile Stimulation Locations  Future investigation will identify the most suitable tactile stimulation locations for our application. We have tested the wrist, the forearm, and the abdomen in this  thesis as tactile stimulation locations. Further comparison between these locations, however, would be required to identify the most suitable location for tactile alerts communication. The most suitable location can be identified by comparing the accuracy, response time, and amount of missed alert in tactile alerts identification at these locations. Identical tactile alert schemes should be used to preserve a fair comparison. Other possible body locations for tactile stimulation will be explored. The tactile display should be placed on a location that preserves high identification  100  rate and rapid response time to tactile alerts identification. Such location, however, should not cause discomfort to the users, nor should it be a burden to the daily work of the anesthesiologists. With reference to the tactile stimulation locations described in Section 2.2.3, the forehead, the foot, and the tongue appeared to be unsuitable for tactile alerts communication. Placing the tactile display at these locations is likely to cause discomfort to the anesthesiologists and create a burden to their daily work. In contrast, the choice of using the human back and the shoulder appear to be worth investigating. Our research group has compared the performance of the tactile display on the human back and the tactile belt display used in Chapter 4. The tactile information conveyed by the tactile display on the back appeared to be more intuitive. This provides us with the incentives to further explore the human back and other locations for tactile stimulation. Future laboratory and clinical evaluation will be conducted to investigate sensory adaptation to tactile alerts, assess users’ comfort after long hours of use, and ultimately, identify the most suitable location for tactile alerts communication.  7.2.4  Clinical Adoption of the Tactile Display  Our elaborate studies on tactile displays have set the stage for translating the tac tile technology for physiological monitoring into the clinical settings, but several issues should be further explored. Firstly, the use of tactile display should not be limited to the operating room. We should study its application in other clinical settings such as the Intensive Care Unit (ICU) and the Emergency Room (ER). Often, clinicians or nurses working in these units are responsible for monitoring the status of several patients simultaneously. Implementing the tactile display in the ICU and ER may improve their awareness of adverse events, hence ensuring a prompt treatment and improving patient’s safety. Secondly, future tactile displays should be tested according to industrial guidelines. These guidelines layout the re quirements for a comprehensive management system for the design and manufacture  101  of medical devices. To allow the adoption of the tactile display in North America, the tactile display is required to pass regulations or guidelines set by, for instance, Food and Drug Administration (FDA), Health Canada, and International Organiza tion for Standardization (ISO). Adoption in European countries such as the United Kingdom would required passing the regulations set by the Medicines and Healthcare products Regulatory Agency (MHRA) and European Union (EU). Passing the standards set by these regulations is the first and mandatory step for clinical adop tion. Thirdly, collaborations with industrial medical device manufacturers should be established. Industrial manufacturers could provide technical support and expertise on improving the usability of the tactile display prototype and reducing the manu facturing cost of the device. Lastly, additional randomized clinical trials are needed to further demonstrate the benefits of a tactile display in improving patients’ safety. Currently, very few similar studies have been performed. More studies investigating other aspects of tactile displays in the clinical environment are therefore required to provide the assurance to the anesthesiologists in adopting this technology. With proven performance and reliability, we foresee tactile displays to be translated into the clinical environment in the future.  In conclusion, results presented in this thesis demonstrate the potential use of tac tile technology to enhance physiological monitoring in the clinical environment. The findings provide directions for the design of a tactile display and the corresponding tactile alert scheme. Our results have established the platform for future transla tion of tactile display to the clinical settings. The ultimate objective is to develop a tactile display that can communicate multiple physiological alerts through the sense of touch, augmenting the monitoring task of the anesthesiologists and improving patient safety.  102  Bibliography [1] Compact oxford english dictionary of current english, 2005. [2] Alles. Information transmission by phantom sensations. IEEE transactions on man-machine systems, 11(1):85, 1970. [3] P. Barralon, C. Ng, G. Dumont, S. K. W. Schwarz, and J. M. Ansermino. Development and evaluation of multidimensional tactons for a wearable tactile display. In MobileHCI ‘07: Proceedings of 9th International Conference on Human Computer Interaction with Mobile Devices and Services, pages 186— 189, Sept 10-12 2007. [4] M. Benali-Khoudja and M. Hafez. Vital: A vibrotactile interface with thermal feedback. In IRCICA: International Scientific Workshop, 2004. [5] M. Benali-Khoudja, M. Hafez, J. M. Alexandre, A. Kheddar, and V. Moreau. Vital: a new low-cost vibro-tactile display system. ICRA ‘0: Proceedings of IEEE International Conference on Robotics and Automation, 1:721—726 Vol.1, April-i May 2004. [6] S. J. Bolanowski. Four channels mediate the mechanical aspects of touch. The Journal of the Acoustical Society of America, 84(5):1680, 1988. [7] 5. A. Brewster and L. M. Brown. Tactons: structured tactile messages for non visual information display. In A UIC ‘04: Proceedings of Fifth Australasian  103  User Interface Conference, pages 15—23, Darlinghurst, Australia, 2004. Aus tralian Computer Society, Inc. [8] 5. A. Brewster and A. King. The design and evaluation of a vibrotactile progress bar. In WHC ‘05: In Proceedings of World Hap tic Conference, pages 499—500, March 2005. [9] C. Brouse, G. Dumont, P. Yang, J. Lim, and J. M. Ansermino. iassist: A soft ware framework for intelligent patient monitoring. In EMBS ‘07: Proceedings of 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pages 3790—3793, Aug. 2007. [10] L. M. Brown, S. A. Brewster, and H. C. Purchase. A first investigation into the effectiveness of tactons. In WHC ‘05: Proceedings of World Haptic Conference, pages 167—176, 2005. [11] L. M. Brown, S. A. Brewster, and H. C. Purchase. Multidimensional tactons for non-visual information presentation in mobile devices. In MobileHCl ‘06: Proceedings of 8th International Conference on Human Computer Interaction with Mobile Devices and Services, pages 231—238, New York, NY, USA, 2006. ACM Press. [12] S. Bryson. Virtual reality in scientific visualization. Communications of the ACM, 39(5):62—71, 1996. [13] G. C. Burdea. Force and touch feedback for Virtual Reality. John Wiley & Sons, Inc., New York, NY, USA, 1996. [14] H. Castle and T. Dobbins. Tactile display technology. Technology And Inno vation, 2004. [15] Categorical. Glossary of terms. Machine learning, 30(2):271, 1998. [16] E. Y. Chen and B. A. Marcus. Exos sup display research and development. 104  [17] H. Y. Chen, J. Santos, M. Graves, K. Kim, and H. Z. Tan. Tactor localization at the wrist. Lecture notes in computer science, 5024:209, 2008. [18] R. A. Cholewiak. Sensory and physiological bases of touch. 1991. In M. A. Heller & W. Schiff (Eds.), The psychology of touch (pp. 23-60). Hilisdale, NJ: Lawrence Erlbaum. [19] R. W. Cholewiak, J. C. Brill, and A. Schwab. Vibrotactile localization on the abdomen: Effects of place and space. Perception é4 Psychophysics, 66(6):970— 987, 2004. [20] R. W. Cholewiak and A. A. Collins. Vibrotactile localization on the arm: Effects of place, space and age. Perception é4 Psychophysics, 65(7):1058—1077, 2003. [21] G. V. Chouvardas, A. N. Miliou, and M. K. Hatalis.  Tactile displays: a  short overview and recent developments. In ICTA ‘05: Proceedings of Fifth International Conference on Technology and Automation. [22] R. J. Christman. Sensory experience. Intext, Scranton; London, 1975. ID: 16304479. [23] J. J. Clinch and H. J. Keselman. Parametric alternatives to the analysis of variance. Journal of educational and behavioral statistics, 7(3):207, 1982. [24] M. B. Cohn, M. Lam, and R. S. Fearing. Tactile feedback for teleoperation. In SPIE ‘93: Proceedings of The International Society for Optical Engineering, pages 240  —  [25] G. G. Cole.  254, 1993. Detectability of onsets versus offsets in the change detection  paradigm. Journal of vision, 3(1):22, 2003.  105  [26] J. Daniels, S. Ford, S. Fels, A. Kushniruk, J. Lim, and J. M. Ansermino. A framework for evaluating usability of clinical monitoring technology. In Society for Technology in Anesthesia, 2007. [27] T. Debus, T. J. Jang, P. Dupont, and R. Howe. Multi-channel vibrotactile dis play for teleoperated assembly. ICRA ‘02: Proceedings of IEEE International Conference on Robotics and Automation, 1:592—597 vol.1, 2002. [281 Dionisio. The virtual touch: Haptic interfaces in virtual environments. Com puters graphics, 21(4) :459, 1997. [29] N. I. Durlach, A. S. Mayor, National Research Council (U.S.). Committee on Virtual Reality Research, Development., and Inc NetLibrary. Virtual real ity scientific and technological challenges, 1995. [30] M. E. H. Eltaib and J. R. Hewit. Tactile sensing technology for minimal access surgery—a review. Mechatronics, 13:1163—1177(15), December 2003. [31] M. Enriquez, 0. Afonin, B. Yager, and K. Maclean.  A pneumatic tactile  alerting system for the driving environment. In P UI ‘01: Proceedings of the 2001 workshop on Perceptive user interfaces, pages 1—7, New York, NY, USA, 2001. ACM. [32] M. Enriquez and K. E. MacLean. Backward and common-onset masking of vibrotactile stimuli. Brain Research Bulletin, 75(6):761—769, 2008. [33] G. Fink, S. Krishnamoorthy, and A. Kanade. Naval crew workload monitoring and visualization. In First Annual Conference on System Integration, March 23-25, 2003. [34] G. A. Finley and A. J. Cohen. Perceived urgency and the anaesthetist: re sponses to common operating room monitor alarms. Anesthesia, 38(8):958—964, 1991. 106  Canadian Journal of  [35] S. Ford, J. Daniels, J. Lim, V. Koval, S. K. W. Schwarz, G. Dumont, and J. M. Ansermino. Touch your patient a human simulation centre based assessment -  of a novel vibrotactile display. In Society for Technology in Anesthesia, 2007. [36] S. Ford, J. Daniels, J. Lim, V. Koval, S. K. W. Schwarz, G. Dumont, and J. M. Ansermino. A novel vibrotactile display to improve the performance of anesthesiologists in a simulated critical incident. Anesth Anaig, 106(11):1182— 1188, April 2008. [37] S. F. Frisken-Gibson. A 64-solenoid, four-level fingertip search display for the blind. IEEE Transactions on Bio-medical Engineering, 34(12):963, 1987. [38] Y. Fujita and S. Hashimoto. Experiments of haptic and tactile display for hu man telecommunication. In RO-MAN ‘99: 8th IEEE International Workshop on Robot and Human Interaction, pages 334—337, 1999. [39] M. Fukumoto and T. Sugimura. Active click: tactile feedback for touch pan els. In CHI ‘01: CHI ‘01 extended abstracts on Human factors in computing systems, pages 121—122, New York, NY, USA, 2001. ACM. [40] A. Gallace, H. Z. Tan, and C. Spence. Tactile change detection. In WHC 2005: First World Haptic Conference, pages 12—16, 2005. [41] F. A. Geldard.  The cutaneous “rabbit”: a perceptual illusion.  Science,  178(57):178, 1972. [42] E. B. Goldstein. Sensation and perception. Thomson/Wadsworth, Belmont, CA, 2004. [43] M. Hafez. Tactile interfaces: technologies, applications and challenges. Vis. Comput., 23(4):267—272, 2007. [44] J. F. Hahn. Low-frequency vibrotactile adaptation. Journal of experimental psychology, 78(4):655, 1968. 107  [45) S. G. Hart and L. Staveland.  Development of nasa-tlx (task load index):  Results of empirical and theoretical research. In Human mental workload, pages 139—183. P.A. Hancock and N. Meshkati (Eds.), Amsterdam: Elsevier, 1988. [46] A. Hem  and M. Brell. contact  -  a vibrotactile display for computer aided  surgery. In WHC ‘07: Proceedings of the Second Joint EuroHaptics Conference and Symposium on Haptic Interfaces for Virtual Environment and Teleoper ator Systems, pages 531—536, Washington, DC, USA, 2007. IEEE Computer Society.  [47] A. Higashiyama. Localization of electrocutaneous stimuli on the fingers and forearm: effects of electrode configuration and body axis. Perception é.4 Psy chophysics, 54(1):108, 1993. [48] H. N. Ho and L. A. Jones. Contribution of thermal cues to material discrimi nation and localization. Perception é1 Psychophysics, 68:118—128(11), 2006. [49] H. N. Ho and L. A. Jones. Development and evaluation of a thermal display for material identification and discrimination. ACM Trans. Appl. Percept., 4(2):13, 2007. [50] B. Hodge. Noise pollution in the operating theatre. Lancet, 335(8694):891, 1990. [51] R. D. Howe, D. A. Kontarinis, and W. J. Peine. Shape memory alloy actuator controller design for tactile displays. Proceedings of the 34th IEEE Conference on Decision and Control, 4:3540—3544 vol.4, Dec 1995. [52) Y. Ikei, K. Wakamatsu, and S. Fukuda. Vibratory tactile display of image based textures.  IEEE Computer Graphics and Applications, 17(6):53—61,  Nov/Dec 1997.  108  [53] S. mo, S. Shimizu, T. Odagawa, M. Sato, M. Takahashi, T. Izumi, and T. Ifukube. A tactile display for presenting quality of materials by chang ing the temperature of skin surface. In Proceedings of 2nd IEEE International Workshop on Robot and Human Communication, pages 220—224, Nov 1993. [54] R. S. Johansson. Tactile sensibility in the human hand: relative and absolute densities of four types of mechanoreceptive units in glabrous skin. The Journal of physiology, 286:283, 1979. [55] L. A. Jones, B. Lockyer, and E. Piateski. Tactile display and vibrotactile pattern recognition on the torso. Advanced robotics, 20(12):1359, 2006. [56] K. A. Kaczmarek. Electrotactile adaptation on the abdomen: preliminary results. IEEE Transactions on Rehabilitation Engineering, 8(4):499—505, Dec 2000. [57] K. A. Kaczmarek and M. E. Tyler. Effect of electrode geometry and intensity control method on comfort of electrotactile stimulation on the tongue.  In  Proceedings of ASME ‘00: Dynamic Systems and Control Division. [58] H. Kajimoto, M. Inami, N. Kawakami, and S. Tachi. Forehead electrotactile display for vision substitution. In Proceedings of Euro-Haptics. [59] H. Kajimoto, M. Inami, N. Kawakami, and S. Tachi.  Smarttouch  -  aug  mentation of skin sensation with electrocutaneous display. In HAPTICS ‘03: Proceedings of 11th Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, pages 40—46, March 2003. [60] H. Kajimoto, N. Kawakami, T. Maeda, and S. Tachi. Electro-tactile display with force feedback. In World Multiconference on Systemics, Cybernetics In formatic, page 95, 2001.  109  [61] H. Kajimoto, N. Kawakami, and S. Tachi. Electro-tactile display with tactile primary color approach. In International Conference on Intelligent Robots and Systems, 2004. [62] H. Kajimoto, N. Kawakami, S. Tachi, and M. Inami. Smarttouch: Electric skin to touch the untouchable. IEEE Computer Graphics and Applications, 24(1):36—43, 2004. [63] P. C. Kam. Noise pollution in the anaesthetic and intensive care environment. Anaesthesia, 49(11):982, 1994. [64] E. R. Kandel, J. H. Schwartz, and T. M. Jessell. Principles of Neural Science. Elsevier, New York, 3rd edition, 2000. [65] G. Keppel and T.D. Wickens. Design and Analysis: A Researcher’s Handbook. Prentice Hall, 1991. [66] I. G. Kesting, B. R. Miller, and C. H. Lockhart.  Auditory alarms during  anesthesia monitoring. Anesthesiology, 69:106—109, 1988. [67] M. P. Kilgard. Anticipated stimuli across skin. Nature, 373(6516):663, 1995. [68] L. T. Kohn, J. Corrigan, and M. S. Donaldson. To err is human: building a safer health system. National Academy Press, Washington, 2000. [69] Kontarinis. Tactile display of vibratory information in teleoperation and vir tual environments. Presence, 4(4):387, 1995. [70] G. Kramer. Auditory Display: Sonification, Audification and Auditory In terfaces. Addison-Wesley Longman Publishing Co., Inc., Boston, MA, USA, 2000. [71] R. H. W. Lam, I. Elhajj, W. J. Li, and L. N. Xi. A bone reaming system with tactile temperature display. In Proceedings of 15th Triennial World Congress of the International Federation of Automatic Control, July 2002. 110  [72] J. C. Lee, P. H. Dietz, D. Leigh, W. S. Yerazunis, and S. E. Hudson. Haptic pen: a tactile feedback stylus for touch screens. In UIST ‘04: Proceedings of the 17th annual ACM symposium on User interface software and technology, pages 291—294, New York, NY, USA, 2004. ACM. [73] V. Levesque, J. Pasquero, and V. Hayward. Braille display by lateral skin deformation with the stress2 tactile transducer. In Proceedings of World Haptics ‘07: Second Joint EuroHaptics Conference and Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, pages 115—120, March 2007. [74] J. R. Lewis. Ibm computer usability satisfaction questionnaires: psychometric evaluation and instructions for use.  mt.  J. Hum.-Comput. Interact., 7(1):57—  78, 1995. [75] R. W. Lindeman, J. L. Sibert, E. Mendez-Mendez, S. Patil, and D. Phifer. Ef fectiveness of directional vibrotactile cuing on a building-clearing task. In CHI ‘05: Proceedings of the SIGCHI conference on Human factors in computing systems, pages 271—280, New York, NY, USA, 2005. ACM. [76] R. G. Loeb. A measure of intraoperative attention to monitor displays. Anes thesia é4 Analgesia, 76:337—341, 1993. [77] R. G. Loeb, B. R. Jones, R. A. Leonard, and K, Behrman. Recognition ac curacy of current operating room alarms. Anesthesia  Analgesia, 75(4):499—  505, 1992. [78] J. M. Loomis. Tactile pattern perception. Perception, 10(1):5, 1981. [79] Nakatani M., H. Kajimoto, D. Sekiguchi, N. Kawakami, and S. Tachi. 3d form display with shape memory alloy. In ICAT2003: International Conference on Artificial Reality and Telexistence, pages 179—184, 2003.  111  [80] Y. Mamiya. Applications of piezoelectric actuator. Special Issue: Electronic Devices, 1(5), 2006. [81] G. A. Miller. The magic number seven, pius or minus two: Some limits on our capacity for processing information. Psychological review, 63(2):81, 1956. [82] K. Momtahan, R. Hetu, and B. Tansley. Audibility and identification of au ditory alarms in the operating room and intensive care unit. Erogonomics, pages 1159—1176, 1993. [83] R. W. Morris and S. R. Montano. Response times to visual and auditory alarms during anaesthesia. Anaesthesia and Intensive Care, 24:682—684, 1996. [84] H. J. Motuisky and R. E. Brown. Detecting outliers when fitting data with nonlinear regression  -  a new method based on robust nonlinear regression and  the false discovery rate. BMC Bioinformatics, 7:123, 2006. [85] G. Moy, C. Wagner, and R. S. Fearing. A compliant tactile display for teletac tion. ICRA ‘00: IEEE International Conference on Robotics and Automation, 4:3409—3415 vol.4, 2000. [86] V. S. Murthy, S. K. Malhotra, I. Bala, and M. Raghunathan. Detrimental effects of noise on anaesthetists. Canadian Journal of Anesthesia, 42(7):608— 611, 1995. [87] G. Ng, P. Barralon, G. Dumont, S. K. W. Schwarz, and J. M. Ansermino. Optimizing the tactile display of physiological information: Vibro-tactile vs. electro-tactile stimuliaton, and forearm or wrist location. In EMBC ‘07: Pro ceedings of p9th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pages 4204—4205, August 23-26 2007. [88] G. Ng, P. Barralon, S. K. W. Schwarz, G. Dumont, and J. M. Ansermino. Evaluation of a tactile display around the waist for physiological monitoring 112  under different clinical workload conditions. In EMBC ‘08: Proceedings of 80th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pages 1288—1291, August 20-24 2008. [89] J. Y. C. Ng, J. C. F. Man, S. Fels, 0. Dumont, and J. M. Ansermino. An evaluation of a vibro-tactile display prototype for physiological monitoring. Anesthesia é4 Analgesia, 101(6):1719—1724, Dec 2005. [90] M. Niwa, Y. Yanagida, H. Noma, K, Hosaka, and Y. Kume. Vibrotactile Apparent Movement by DC Motors and Voice-coil Tactors.  In ICAT ‘04:  Proceedings of 14th International Conference on Artificial Reality and Telex istence, Seoul, Korea, pages 126—131, 2004. [91] T. Nojima and K. Funabiki. Cockpit display using tactile sensation. World Haptics ‘05: First Joint Eurohaptics Conference and Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, pages 501—502, March 2005. [92] T. Ohtsuka, A. Furuse, T. Kohno, J. Nakajima, K. Yagyu, and S. Omata. Application of a New Tactile Sensor to Thoracoscopic Surgery: Experimental and Clinical Study. The Annals of Thoracic Surgery, 60(3):610—613, 1995. [93] M. V. Ottermo, 0. Stavdahl, and T. A. Johansen.  Palpation instrument  for augmented minimally invasive surgery. In IROS ‘04: Proceedings of In ternational Conference on Intelligent Robots and Systems, volume 4, pages 3960—3964, 2004. [94] J. Pasquero and V. Hayward. Stress: A practical tactile display system with one millimeter spatial resolution and 700 hz refresh rate. In EuroHaptics p008, Dublin, Ireland, 2003. [95] E. Piateski and L. Jones. Vibrotactile pattern recognition on the arm and torso. World Haptics ‘05: First Joint Eurohaptics Conference and Symposium 113  on Haptic Interfaces for Virtual Environment and Teleoperator Systems, pages 90—95, March 2005. [96] C. D. Ray.  Noise pollution in the operating room: a hazard to surgeons,  personnel, and patients. Journal of Spinal Disorders, 5(4):485, 1992. [97] E. Sampaio, S. Mans, and P. Bach y Rita. Brain plasticity: ‘visual’ acuity of blind persons via the tongue. Brain Research, 908(2):204—207, 7/27 2001. [98] P. Sanderson, J. Crawford, A. Savill, M. Watson, and W. Russell. Visual and auditory attention in patient monitoring: a formative analysis. Cognition Technology and Work, 6(3):172, 2004. [99] P. Sanderson and F. J. Seagull. Cognitive ergonomics of information tech nology in critical care: contexts and modalities for alarm interpretation. In Proceedings of International Workplace Health and Safety Forum and 33rd Confernce of the Ergonomics Society of Australia, pages 43—52, 1997. [100] P. M. Sanderson, M. 0. Watson, and J. W. Russell. Advanced Patient Mon itoring Displays: Tools for Continuous Informing. Anesthesia & Analgesia, 101(1):161—168, 2005. [101] K. Sato, E. Igarashi, and M. Kimura. Development of non-constrained master arm with tactile feedback device. In ICAR ‘91: Proceedings of Fifth Interna tional Conference on Advanced Robotics, volume 1, pages 334—338, Jun 1991. [102] 5. 5. Sawilowsky and R. C. Blair. A more realistic look at the robustness and type ii error properties of the t test to departures from population normality. Psychological bulletin, 111(2):352, 1992. [103] R. Scott. Medical device alarm standards-the current position. Medical Equip ment Alarms The need, The Standards, The Evidence (Ref. No. 1998/432), lEE Colloquium On, pages 3/1—3/6, Oct 1998. 114  [1041 F. J. Seagull, C. D. Wickens, and R. G. Loeb. When is less more attention  and workload in auditory, visual, and redundant patient-monitoring condi tions. Proceedings of Human Factors and Ergonomics Society Annual Meeting, 45:1395—1399(5), 2001. [1051 C. Sherrick and R. Cholewiak, Cutaneous sensitivity. In Handbook of Percep tion and Human Performance. Wiley, New York, 1986. [106] D. C. Sinclair. Mechanisms of Cutaneous Sensation. Oxford University Press,  New York, 2nd edition, 1981. [107] Canadian Anesthesiologists’ Society. Guidelines to the practice of anesthesia. Supplement to the Canadian Journal of Anesthesia, 54:12, 2007. [108] R. R. Sokal and F. J. Rohlf. Biometry. Freeman, San Francisco, 2nd edition, 1981.  [109] K. M. Stanney and Inc NetLibrary. Handbook of virtual environments design, implementation, and applications, 2002. [110] A. M. Stölle, R. Hözl, D. Kleinböl, A. Mrsic, and H.Z. Tan. Measuring poing localization errors in spatiotemporal tactile stimulus patterns. In EuroHaptics, pages 512—515, 2004. [111] M. K. Sykes, M. D. Vickers, C. J. Hull, and M. K. Sykes.  Principles of  measurement and monitoring in anaesthesia and intensive care.  Blackwell  Scientific Publications St Louis, Mo.: Distributors, USA, Mosby-Year Book, Oxford ; Boston, 1991. [112] H. Z. Tan, N. I. Durlach, C. M. Reed, and W. M. Rabinowitz. Information transmission with a multifinger tactual display. Perception é4 Psychophysics, 61(6):993—1008, Aug 1999.  115  [113] W. Y. Tan. Sampling distributions and robustness of t, f, and variance-ratio in two samples and anova models with respect to departure from normality. Communications in Statistics, 11:2485, 1982. [114] H. Tang and D. J. Beebe.  Design and microfabrication of a flexible oral  electrotactile display. Journal of Microelectromechanical Systems, 12(1) :29— 36, Feb 2003. [115] Taylor. A sixty-four element tactile display using shape memory alloy wires. Displays, 18(3):163, 1998. [116] A. Toney, L. Dunne, B. H. Thomas, and S. P. Ashdown. A shoulder pad insert vibrotactile display.  In ISWC ‘03: Proceedings of 7th IEEE International  Symposium on Wearable Computers, page 35, Washington, DC, USA, 2003. IEEE Computer Society. [117] K. K. Tremper. Pulse oximetry. Anesthesiology, 70(1):98, 1989. [118] K. Tsukada and M. Yasumura. Active belt: Belt-type wearable tactile display for directional navigation. Special Issue: Research on Interaction: Theories, Technologies, Applications and Evaluations, 44(11):2649—2658, 2003. [119] W. R. Uttal. Inhibitory interaction of responses to electrical stimuli in the fingers. Journal of Comparative and Physiological Psychology, 56:47—51, 1960. [120] J. B. F. van Erp.  Tactile navigation display.  In Proceedings of the First  International Workshop on Haptic Human-Computer Interaction, pages 165— 173, London, UK, 2001. Springer-Verlag. [121] J. B. F. van Erp and H. A. H. C. van Veen. Vibrotactile in-vehicle navigation system. Transportation Research Part F: Traffic Psychology and Behaviour, 7(4-5):247—256, 0 2004,  116  [122] J. B. F. van Erp, H. A. H. C. van Veen, C. Jansen, and T. Dobbins. Waypoint navigation with a vibrotactile waist belt. ACM Transactions on Applied Perception, 2(2):106—117, 2005. [123] J. B. F. van Erp, J. A. Veltman, H. A. H. C. van Veen, and A. B. Oving. Tactile torso display as countermeasure to reduce night vision goggles induced drift. In NATO RTO HFM Panel Symposium Spatial Disorientation in Military Vehicles: Causes, Consequences and Cures, 2002. [124] R. T. Verrillo. Vibrotactile masking: effects of one- and two-site stimulation. Perception  Psychophysics, 33(4):379, 1983.  [125] F. Vidal-Verdü and R. Navas-González. Thermopneumatic actuator for tactile displays. In DCIS 2003: Proceedings of 18th Conference on Design of Circuits and Integrated System, pages 629—633, 2003. [126] D. L. Weber and D. M. Green.  Temporal factors and suppression effects  in backward and forward masking. The Journal of the Acoustical Society of America, 64(5):1392—1399, 1978. [127] M. B. Weinger. Ergonomic and human factors affecting anesthetic vigilance and monitoring performance in the operating room environment. Anesthesi ology, 73(5):995, 1990. [128] S. Weinstein. Intensive and extensive aspects of tactile sensitivity as a function of body part, sex, and laterality. In The skin senses, 1968. Edited by D. R. Kenshalo. Springfield, Ill, Charles C Thomas, pages 195—222. [129] J. M. Weisenberger. The transmission of phoneme-level information by mul tichannel tactile speech perception aids. Ear and Hearing, 16(4):392, 1995. [130] P. Wellman, W. Peine, G. Favalora, and R. Howe. Mechanical design and control of a high-bandwidth shape memory alloy tactile display. In Proceedings of International Symposium on Experimental Robotics, June 1997. 117  [1311 P. S. Weilman, E. P. Dalton, D. Krag, K. A. Kern, and R. D. Howe. Tac tile Imaging of Breast Masses: First Clinical Report. Archives of Surgery, 136(2):204—208, 2001. [132] J. J. Wertsch, P. Bach y Rita, M. B. Price, J. Harris, and J. Loftsgaarden. Development of a sensory substitution system for the insensate foot. Journal of Rehabilitation Research and Development, 25:269—270. [133] Y. Xiao, F. Mackenzie, J. Seagull, and M. Jaberi. Managing the monitors: an analysis of alarm silencing activities during an anesthetic procedure. In Proceedings of 44th Annual Meeting of the Human Factors and Ergonomics Society, pages 250—253. Human Factors and Ergonomics Society, 2000. [134] A. Yamamoto, B. Cros, H. Hashimoto, and T. Higuchi. Control of thermal tactile display based on prediction of contact temperature.  In ICRA ‘04:  Proceedings of IEEE International Conference onRobotics and Automation, volume 2, pages 1536—1541 Vol.2, 26-May 1, 2004. [135] Y. Yanagida, M. Kakita, R. W. Lindeman, Y. Kume, and N. Tetsutani. Vi brotactile letter reading using a low-resolution tactor array. In HAPTICS ‘04: Proceedings of 12th International Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems, pages 400—406, March 2004. [136] P. Yang, G. Dumont, and J. M. Ansermino. An adaptive cusum test based on a hidden semi-markov model for change detection in non-invasive mean blood pressure trend. EMBC ‘06: Proceedings of 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pages 33395  —  3398, Aug. 2006.  [137] P. Yang, G. A. Dumont, and J. M. Ansermino. Adaptive change detection in heart rate trend monitoring in anesthetized children. IEEE Transactions on Biomedical Engineering, 53(11) :2211—2219, 2006. 118  [1381 T. H. Yang, J. S. Lee, S. S. Lee, S. Y. Kim, and D. S. Kwon. Conceptual design of new micro-actuator for tactile display. In ICCAS ‘07: Proceedings of International Conference on Control, Automation and Systems, pages 1306— 1309, Oct. 2007. [139] J. Yanof, C. Bauer, and B. Wood. Tactile feedback and display system for ct-guided, robot-assisted percutaneous procedures. In CARS ‘04: Proceed ings of the 18th International Congress and Exhibition on Computer Assisted Radiology and Surgery, volume 1268, pages 521—526, 2004. [140] J. T. W. Yeow and B. Cheung. An integrated tactile and visual display motion simulator for air and land application. In Proceedings of IEEE International Conference on Mechatronics and Automation, volume 1, pages 298—302 Vol. 1, July-i Aug. 2005. [141] E. Yeung, A. Boothroyd, and C. Redmond. A wearable multichannel tactile display of voice fundamental frequency, Ear and Hearing, 9(6) :342, 1988.  119  Appendix A  Statement of Co-authorship I, Yee Lam Ginna Ng, have written all chapters in this thesis under the supervision of Prof. Guy Dumont and Dr. Mark Ansermino. Despite the close collaboration with Dr.  Pierre Barralon in the research  project, I contributed in all aspects, from the experimental design to the statistical analysis, of the four user studies presented in this thesis.  Details regarding my  contributions on each study are described as follows: User Study #1: Evaluation of Vibrotactile and Electrotactile Stimulation, and Vibrotactile Localization on Forearm and Wrist Contributions include designing the tactile alert scheme that was used to convey tactile information to the tactile display users and developing the three physical tactile display prototypes used in the study, namely, vibrotactile display on the forearm (VF), vibrotactile display on the wrist (VW), and electrotactile display on the forearm (EF). In addition to planning with Dr. Barralon the study procedures, I was responsible for writing the clinical trial protocol which was then submitted to the IJBC Clinical Research Ethics Board for ethics approval. Prior to writing the clinical trial protocol, I conducted a pilot study with eight non-medically trained participants to compare the accuracy and response time 120  on tactile alert identification between the same VF and EF prototypes used in the actual study. Apart from fulfilling a course requirement, the results of this pilot study were included in the protocol for sample size estimation. Other contributions include recruiting and scheduling study appointments with 30 study participants on the UBC campus, The study was completed in ap proximately two months and was conducted together by Dr. Barralon and myself. In addition to performing the statistical analysis with Dr. Barralon, I submitted, as the first author, a conference manuscript to the 29th Annual International Con ference of the IEEE Engineering in Medicine and Biology Society (EMBC) in Lyon, France [87]. I was responsible for presenting the work in an oral session at the con ference. The results were also presented at the domestic workshop TechMed 2006. Additional analysis presented in this thesis was conducted by myself and the find ings will be included in a journal publication which will be submitted in the near future. User Study #2: Designing a Tactile Alert Scheme with Multidimensional Tactons In addition to initiating the idea of incorporating the concept of Tactons [7] into the design of the tactile alert scheme and selecting the abdomen as the stimulation location, I was responsible for planning the details of the study with Dr. Barralon. This included designing the study procedures and choosing the Tacton components and their combinations for the tactile alert scheme. Regarding other individual contributions, I wrote the clinical trial protocol of the study, which was submitted for ethics approval to UBC Clinical Research Ethics Board. Furthermore, I promoted the study to peers and students of UBC through verbal correspondence and online advertisement on HCI©UBC, an online experiment management system operated by the university. I was also responsible for coordinating and scheduling study appointments with 30 study participants for  121  this study which has lasted for approximately 1.5 months. Conducting the actual study together, Dr. Barralon and I also worked to gether on the statistical analysis for a conference manuscript published in the pro ceedings of the 9th International Conference on Human Computer Interaction with Mobile Devices and Services (MobileHCI) in Singapore [3]. I was the second author of the manuscript. In addition, I prepared and presented a poster for this study at Child and Family Research Institute Student Research Forum 2007. Additional sta tistical analysis was conducted by myself and presented in this thesis. The findings will be included in another journal publication which will be submitted in the near future. User Study #3: Perception of Rhythm-based Tactile Alerts on the Ab domen My contribution focused on planning the study procedures, introducing the encod ing parameters to describe the tactile alert scheme, and designing the tactile alerts for the four tactile alert schemes used in the study with Dr. Barralon. I was also responsible for promoting the study through verbal correspondence, online adver tisement on UBCHCI, and departmental emails, as well as communicating and scheduling study appointments with 64 study participants. The clinical trial protocol of this study, which was submitted to UBC Clin ical Research Ethics Board for ethics approval, was written by myself. I was also responsible for the statistical analysis presented in this thesis. The study took ap proximately two months to complete and was conducted together by Dr. Barralon and myself. I was the second author of a journal publication that has been submitted to IEEE Transaction on System, Man and Cybernetics (Part A).  122  User Study #4: Performance of Tactile Belt Display under Simulated Low and High Clinical Workload Conditions Collaboration with Dr. Barralon on this study was limited to the planning of the study procedures and the design of the tactile alert scheme. Otherwise, I was re sponsible for all other aspects of this study.  This included writing the clinical  trial protocol that was submitted to UBC Clinical Research Ethics Board, British Columbia Children’s Hospital, and St. Paul’s Hospital for ethics approval, recruiting the study participants, and performing the post-study statistical analysis. Since the participants were certified anesthesiologists and anesthesia resi dents, I promoted the study at the Department of Anesthesia Research Meeting at British Columbia Children’s Hospital. To recruit a total of 30 participants, I visited the operating rooms at British Columbia Children’s Hospital and St. Paul’s Hospital, approached all anesthesiologists and anesthesia residents to identify po tential study participants, and communicated with the anesthesiologist-in-charge to schedule study appointments. Furthermore, I conducted a pilot study to validate the change in workload between the two stimulated clinical scenarios used in the actual study. In addition to modifying the Visual Basic program, initially prepared by Dr. Barralon, for the pilot study, I was responsible for communicating, scheduling study appointments, and conducting the pilot study with 12 non-medically trained study participants from British Columbia Children’s Hospital and Electrical and Computer Engineering for Medicine Lab. I also performed the statistical analysis on the data collected from the pilot study. Conducting the subsequent statistical analysis collected from the actual study, I presented, as the first author, the interim results in a conference manuscript pub lished in the proceedings of the 30th Annual IEEE Engineering in Medicine and Biology Society (EMBC) in Vancouver, Canada [88]. Apart from preparing the poster for the conference, I presented the work at the Child and Family Research  123  Institute Student Research Forum 2008. The results of the completed study were also presented at the Research Grand Rounds at the Department of Anesthesia in St. Paul’s Hospital. These findings will be included in a journal publication that will be submitted in the near future.  124  

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.24.1-0067059/manifest

Comment

Related Items