Next RISE-MICCAI Journal Club: Nov 15th
Wednesday 5th November 2025
Join the next RISE-MICCAI Journal Club Session:
Agentic AI for radiology education
Presenting author: Akash Awasthi, University of Houston
Saturday, November 15, 2025
12:00 pm EST / 6:00 pm CET
Abstract: Radiology education requires trainees to develop both perceptual and interpretive expertise, yet this learning process is often limited by the scarcity of expert mentorship and the absence of personalized feedback. Many diagnostic errors stem from subtle perceptual lapses—missed fixations, brief dwell times, or overlooked abnormalities—that current AI systems fail to capture or explain.
In this talk, I will present our recent work on intelligent agent-based frameworks that integrate gaze behavior, diagnostic reasoning, and adaptive feedback to advance radiology training. The first framework, Structural Chain of Thoughts (SCoT), structures gaze and report data into a thought graph, enabling Large Language and Multimodal Models (LLMs/LMMs) to detect fine-grained perceptual and interpretive discrepancies between experts and trainees, and to provide interpretable, context-aware feedback. Building on this, MAARTA extends SCoT into a multi-agent system that dynamically recruits reasoning agents based on error complexity. Using Chain-of-Thought prompting and specialized Perceptual Error Teacher (PET) agents, MAARTA identifies missed findings, analyzes perceptual differences, and delivers tailored educational insights. Together, SCoT and MAARTA represent a new direction toward adaptive, interpretable, and scalable AI tutors that bridge the gap between novice and expert diagnostic reasoning—offering a glimpse into the future of personalized, multimodal medical education.
Bio: Akash Awasthi is a Machine learning Researcher in Electrical and Computer Engineering at the University of Houston. His research focuses on developing collaborative AI systems using Large Multimodal Models (LMMs) for healthcare education and decision support. His work spans artificial intelligence, computer vision, and interdisciplinary scientific applications, with publications in leading AI and medical imaging venues.