University of British Columbia
Towards the feasibility of prototypical machine learning for mitral stenosis detection in echocardiography
Abstract
dc:descriptionMitral stenosis, or the pathological narrowing of the mitral valve orifice, is typically detected and monitored using echocardiography. Echocardiography allows clinicians to assess the morphology of the mitral valve apparatus and analyze the hemodynamic consequences of the narrowed orifice through the use of Doppler. However, as with most ultrasound modalities, echocardiographic image acquisition and interpretation are highly operator-dependent, making it challenging to obtain diagnostic-quality images without specialized training. While numerous machine learning approaches have been developed to address these limitations and automate the analysis of echocardiogram videos for various valvular pathologies, the application of these methods to mitral stenosis presents unique challenges. Mitral stenosis is relatively uncommon, resulting in limited availability of annotated training data, which complicates the development of robust models. Furthermore, many existing machine learning frameworks function as ”black-box” systems, offering little interpretability and transparency regarding their decision-making process—–an important barrier to clinician trust and clinical integration. This thesis addresses these challenges by introducing ProtoMSNet, a prototypical machine learning model designed to identify mitral stenosis using echocardiographic views that are easily acquired. ProtoMSNet offers explainable decisions through the use of informative prototypes, enhancing trust and interpretability. To mitigate data scarcity, the thesis further explores the use of an etiology-based transfer learning strategy, in which ProtoMSNet is pre-trained on mitral valve leaflet thickness and mitral annular calcification—features strongly associated with the most common causes of mitral stenosis: rheumatic valve disease and calcific valve disease. Our results demonstrate that ProtoMSNet not only accurately distinguishes cases of mitral stenosis, but also generates representative prototypes that highlight the diverse etiologies of this condition.
Degree
thesis:*- Name thesis:degree_name
- Master of Applied Science - MASc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Biomedical Engineering
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Goco, Jamie Alexis
- Contributors dc:contributor
-
- Abolmaesumi, Purang
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/2429/92628