University of British Columbia
Improving robustness of deep learning models for processing ultrasound volumes for assessing developmental dysplasia of the hip
Abstract
dc:descriptionDevelopmental Dysplasia of the Hip (DDH) is a painful orthopaedic malformation diagnosed at birth in 1-3% of all newborns. Left untreated, DDH can lead to significant morbidity including long term disability. Currently the condition is clinically diagnosed using 2D ultrasound (US) imaging acquired between 0-6 months of age. DDH metrics are manually extracted by highly trained radiologists through manual measurements of relevant anatomy from the 2D US data, which remains a time consuming and highly error prone process. Recently, it was shown that combining 3D US imaging with deep learning (DL)-based automated diagnostic tools may significantly improve accuracy and reduce variability in measuring DDH metrics. However, robustness of current techniques remains insufficient for reliable deployment into real life clinical workflows. In this thesis, we present a quantitative robustness evaluation of state-of-the-art (SOTA) DL models in bone segmentation for 3D US and demonstrate examples of failed or implausible segmentations with SOTA models under common data variations, e.g., small changes in image resolution or anatomical field of view (FOV) from those encountered in the training data. We propose a 3D extension of the Seg- Former architecture, a lightweight transformer-based model with hierarchically structured encoders producing multi-scale features, which we show to concurrently improve accuracy and robustness. Specifically we show an increase in the 3% Dice score performance over the previous SOTA models for 3D US segmentation. To allow researchers, collaborators, clinicians, and doctors access to our DL models, we develop a prototype web-based application that will allow users to upload three dimensional US data and visualize their data before eventually selecting from various DL models to run on their data. The DL models will run in the background segmenting out the hip anatomical structures and return the calculated DDH metrics as well as relevant visualization of the segmentation and a 3D rendered mesh of the hip from the segmentation. We also investigate the use of learnable Gabor Filter Banks as a preprocessing layer in DL models to mimic the human visual system.
Degree
thesis:*- Name thesis:degree_name
- Master of Applied Science - MASc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Electrical and Computer Engineering
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Hers, Benjamin
- Contributors dc:contributor
-
- Abugharbieh, Rafeef
- Hodgson, Antony J.
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/88098