Ahmad P. Tafti
University of Pittsburgh
Title: AI and Clinicians-in-the-Loop: Rethinking AI Explainability in Medical Imaging
Abstract:
Artificial intelligence (AI) has achieved remarkable progress in medical imaging, yet explainability remains a major barrier to clinical trust and adoption. Most current explainable AI approaches mainly focus on model-centric techniques, such as saliency maps and attention visualization, offering insight into model behavior but often failing to align with how clinicians actually reason and make decisions. This talk argues that explainability in medical imaging should not be treated as an AI-only problem. Instead, meaningful explainability emerges through collaboration between AI systems and clinical experts. By placing clinicians in the loop, explanations become contextual, interactive, and clinically actionable rather than merely visual. Through real-world examples and current research trends, the session explores the strengths and limitations of state-of-the-art explainable methods and discusses how human-centered workflows can improve trust, transparency, and clinical usability in medical imaging AI systems.
About the Speaker:
