Open Collections

Shi, Congzhen

University of British Columbia

The rapid advancement of foundation models in medical imaging represents a significant step toward improving diagnostic accuracy and personalized treatment. However, the deployment of these models in healthcare requires a rigorous examination of their trustworthiness, which includes aspects such as privacy, robustness, reliability, explainability, and fairness. Despite their potential, the current literature on foundation models in medical imaging reveals notable gaps in addressing trustworthiness, particularly regarding segmentation, medical report generation, medical question answering (Q&A), and disease diagnosis. To address this, we conduct a comprehensive survey that analyzes existing research, identifies challenges, and proposes research directions to build more trustworthy foundation models. Our review highlights the critical need for a balanced approach that advances innovation while ensuring ethical and equitable healthcare delivery. Building on this, we explore the explainability of machine learning models in predicting antimicrobial use within electronic health records (EHRs). Using a dataset of 10,816 patients from the Vancouver Coastal Health Community (2019–2022), we developed seven machine learning models to predict antimicrobial use as a proxy for wound infection risk. The SHAP method was employed to interpret model predictions, identifying key risk factors such as specific wound bed characteristics and high frequency of dressing change. The findings demonstrate the value of explainable AI in enhancing clinical decision-making and trustworthiness in healthcare. Collectively, this thesis bridges the critical gap between trustworthiness and explainability in AI applications for healthcare. The work contributes to a holistic understanding of deploying AI responsibly while addressing ethical and clinical considerations.

Text

2025-04-14

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/90655

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/