UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Towards a robust estimation of ejection fraction : a deep uncertainty aware approach Kazemi Esfeh, Mohammad Mahdi


Ejection fraction is a widely-used and critical index of cardiac health. It measures the efficacy of the cyclic contraction of the ventricles and the outward pumpage of blood through the arteries. Timely and robust evaluation of ejection fraction is essential, as reduced ejection fraction indicates dysfunction in blood delivery during the ventricular systole, and is associated with a number of cardiac and non-cardiac risk factors and mortality-related outcomes. Automated reliable ejection fraction estimation in echocardiography has proven challenging due to low and variable image quality, and limited amounts of data for training data-driven algorithms, which delays the integration of the technologies in the clinical workflow. Deep learning has shown state-of-the-art performance in many learning tasks especially in learning from image and video datasets. While deep learning models give promising results in these fields, they are usually over-confident about their outputs and predictions. However, in many applications like the ones related to human health and safety, a well-calibrated and reliable uncertainty estimation is required. In this thesis, we review the most important results in the literature of uncertainty estimation in deep learning and then, propose multiple Bayesian and non-Bayesian deep models to estimate the ejection fraction from echocardiography data along with the epistemic and aleatoric uncertainties associated with these estimations. Finally, we evaluate these models by training and testing them on a publicly available dataset, and by making a side-by-side comparison of them with their deterministic counterparts. Our results show the feasibility of those methods to be deployed in healthcare applications. Also based on the presented rationale and results, we believe that the proposed approach can further be thought of as a generic approach for a more robust evaluation of critical clinical indices.

Item Citations and Data


Attribution-NonCommercial-NoDerivatives 4.0 International