Back to search

University of British Columbia

Towards a more robust machine learning framework for computer-assisted echocardiography

Abstract

dc:description

Heart disease is one of the foremost causes of mortality worldwide. Echocardiography (echo) is a commonly used modality to study the heart's structure and function as it is non-invasive and cost-effective. In recent years, the emergence of readily accessible ultrasound (US) devices, namely point-of-care ultrasound (POCUS), has accelerated the widespread uptake of echo. However, robust echo acquisition and interpretation is tedious, and its efficacy depends on the skills of expert physicians. In this thesis, we investigate machine learning solutions to promote reliable computer-assisted echo examination, improving interpretation accuracy, diagnostic throughput, and test-retest reliability. The main tackled challenges include considering temporal data dependencies, label sparsity, multi-task learning, low-quality noisy nature of echo, and noisy clinical labels with high inter- and intra-observer variability. We present a deep spatio-temporal model integrating recurrent fully convolutional neural networks and optical flow estimation maps to accurately track the left ventricle in echo video clips. We also propose a semi-supervised learning algorithm to leverage unlabeled data to improve the performance of machine learning methods. Moreover, we present a computationally efficient mobile framework for accurate left ventricular ejection fraction estimation. The proposed mobile application runs in real time on an Android smartphone with a connection to a POCUS or cart-based ultrasound device. We also suggest adapting conditional generative adversarial network (cGAN) architectures to improve the quality of echo data. Further, we investigate predictive uncertainties via Bayesian deep learning to sustain the robust deployment of the developed methodologies.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy - PhD
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Electrical and Computer Engineering
Grantor dc:publisher
University of British Columbia
Year dc:date
2021

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Jafari, Mohammad H.
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/2429/78505

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Jafari, Mohammad H.. Towards a more robust machine learning framework for computer-assisted echocardiography. doctoral thesis, University of British Columbia, 2021. http://hdl.handle.net/2429/78505