CS Events

PhD Defense Towards Generalist Medical Artificial Intelligence: Model, Data, Knowledge and Beyond
Export Event Save iCal Send to Google Calendar Send to Yahoo Calendar Send to Outlook Live Send to Microsoft Outlook Save iCal Preserve formatting in description (only supported in some calendar applications) Close Wednesday, November 13, 2024, 03:15pm - 04:30pm

Speaker: Yunhe Gao

Location : CoRE 301

Committee: 

Professor Dimitris Metaxas (chair)

Assistant Yongfeng Zhang

Associate Professor Konstantinos Michmizos

Professor James Duncan

Event Type: PhD Defense

Abstract: Artificial intelligence has demonstrated remarkable capabilities in processing and understanding diverse forms of data - analyzing images, comprehending language, recognizing speech, and interpreting complex patterns across multiple modalities. Despite these broad advances, healthcare remains a uniquely challenging domain where conventional AI approaches often fall short of clinical requirements. While current medical AI systems excel at specific tasks, they lack the generalist capabilities of human doctors who can interpret subtle patterns across diverse imaging modalities, reason about complex medical conditions, and generate comprehensive diagnostic reports. This gap is particularly evident given healthcare's distinct challenges: the need to interpret intricate anatomical patterns where errors can have serious clinical consequences, the scarcity of high-quality medical data due to privacy concerns and annotation costs, and the critical requirement to incorporate centuries of accumulated medical knowledge. This dissertation addresses these fundamental challenges through a comprehensive framework built on three interconnected pillars: Model, Data, and Knowledge. Our research evolves medical AI from narrow, task-specific solutions toward generalist models capable of handling diverse imaging modalities and diagnostic tasks simultaneously. By developing innovative architectural approaches, efficient learning strategies for limited data scenarios, and methods for integrating established medical expertise, we bridge the gap between pure data-driven approaches and clinical practice. The results demonstrate significant improvements in model generalization, data efficiency, and clinical relevance, marking a crucial step toward more reliable and versatile medical AI systems that can match the breadth of human doctors' capabilities.

: 

Contact  Professor Dimitris Metaxas (Chair)

Zoom Link: https://rutgers.zoom.us/j/8154809666?pwd=ZjZGWmpJdkNYODhLUnBPb1VhcXJGdz09&omn=91816714466