- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Fully automatic 3D ultrasound techniques for improving...
Open Collections
UBC Theses and Dissertations
UBC Theses and Dissertations
Fully automatic 3D ultrasound techniques for improving diagnosis of developmental dysplasia of the hip in pediatric patients : classifying scan adequacy and quantifying dynamic assessment Paserin, Olivia
Abstract
Developmental dysplasia of the hip (DDH) is the most common pediatric hip disorder, representing a spectrum of hip instabilities from mild to complete dislocation. Routine DDH clinical examinations consist of two parts: static assessment, for evaluating acetabular morphologies with ultrasound (US), and dynamic assessment, for detecting abnormal hip instabilities by applying stress to the joint and feeling the resulting movement. Several recent works have shown that 3D US computer-aided methods significantly reduce dysplasia metrics’ variability by 70% compared to standard 2D approaches. However, identifying adequate diagnostic US volumes is a challenging task and dynamic assessment has been shown to be relatively unreliable. In this thesis, we propose automated techniques to classify 3D US scan adequacy and a repeatable method for quantifying femoral head displacement observed during dynamic assessment with 3D US. To automatically classify scan adequacy, we developed and evaluated three near real-time deep learning techniques that build upon each other from individual slice by slice categorization with a convolutional neural network to long range inter-slice analysis with a recurrent neural network. Our contributions include developing effective criteria that defines the features required for DDH diagnosis in an adequate 3D US volume, proposing an efficient architecture for robust classification, and validating our model's agreement with expert radiologist labels. We achieved 80% per volume accuracy on a test set of 20 difficult to interpret volumes and a runtime of two seconds. To quantify dynamic assessment, we propose an automatic method of calculating the observed degree of movement through a novel 3D femoral head coverage displacement metric. We designed and conducted a clinical study to record dynamic assessment manoeuvres with 3D US on a cohort of 40 pediatric patients. We evaluated our 3D femoral head coverage displacement metric and found a good degree of repeatability with a test-retest ICC measure of 0.70 (95% CI: 0.51 to 0.83, p<0.001). Ultimately, this work is to be integrated into a complete automated 3D US tool, which may lead to a more standardized and universal DDH assessment compared to current standard practice.
Item Metadata
Title |
Fully automatic 3D ultrasound techniques for improving diagnosis of developmental dysplasia of the hip in pediatric patients : classifying scan adequacy and quantifying dynamic assessment
|
Creator | |
Publisher |
University of British Columbia
|
Date Issued |
2018
|
Description |
Developmental dysplasia of the hip (DDH) is the most common pediatric hip disorder, representing a spectrum of hip instabilities from mild to complete dislocation. Routine DDH clinical examinations consist of two parts: static assessment, for evaluating acetabular morphologies with ultrasound (US), and dynamic assessment, for detecting abnormal hip instabilities by applying stress to the joint and feeling the resulting movement. Several recent works have shown that 3D US computer-aided methods significantly reduce dysplasia metrics’ variability by 70% compared to standard 2D approaches. However, identifying adequate diagnostic US volumes is a challenging task and dynamic assessment has been shown to be relatively unreliable. In this thesis, we propose automated techniques to classify 3D US scan adequacy and a repeatable method for quantifying femoral head displacement observed during dynamic assessment with 3D US.
To automatically classify scan adequacy, we developed and evaluated three near real-time deep learning techniques that build upon each other from individual slice by slice categorization with a convolutional neural network to long range inter-slice analysis with a recurrent neural network. Our contributions include developing effective criteria that defines the features required for DDH diagnosis in an adequate 3D US volume, proposing an efficient architecture for robust classification, and validating our model's agreement with expert radiologist labels. We achieved 80% per volume accuracy on a test set of 20 difficult to interpret volumes and a runtime of two seconds.
To quantify dynamic assessment, we propose an automatic method of calculating the observed degree of movement through a novel 3D femoral head coverage displacement metric. We designed and conducted a clinical study to record dynamic assessment manoeuvres with 3D US on a cohort of 40 pediatric patients. We evaluated our 3D femoral head coverage displacement metric and found a good degree of repeatability with a test-retest ICC measure of 0.70 (95% CI: 0.51 to 0.83, p<0.001).
Ultimately, this work is to be integrated into a complete automated 3D US tool, which may lead to a more standardized and universal DDH assessment compared to current standard practice.
|
Genre | |
Type | |
Language |
eng
|
Date Available |
2019-01-07
|
Provider |
Vancouver : University of British Columbia Library
|
Rights |
Attribution-NonCommercial-NoDerivatives 4.0 International
|
DOI |
10.14288/1.0375892
|
URI | |
Degree | |
Program | |
Affiliation | |
Degree Grantor |
University of British Columbia
|
Graduation Date |
2019-02
|
Campus | |
Scholarly Level |
Graduate
|
Rights URI | |
Aggregated Source Repository |
DSpace
|
Item Media
Item Citations and Data
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International