Towards Trustworthy AI in Medical Image Analysis

Abstract

Despite advancements in AI models for medical image analysis, ensuring their trustworthiness remains a challenge. Accuracy, interpretability, robustness, and generalizability are critical for gaining the confi-dence of healthcare professionals and other stakeholders. This talk focuses on AI model interpretability and generalizability, key aspects of trustworthiness, by exploring the following core questions: 1) Are explainable AI methods reliable, and can we quantitatively measure that? 2) Can we predict when a model might fail to generalize, and how can we enhance its generalizability? 3) How can pre-trained medical Vision-Language Models be adapted to new diseases with limited data? These questions are pivotal in advancing trustworthy AI applications in healthcare.

For more info, please follow this link.