ASGE Postgraduate Course at ACG: Innovative Practices for the Endoscopist of Today, Tomorrow and Beyond | October 2023-13

13_Berzin State of the Art Directions in AI

Pdf Summary

Dr. Tyler Berzin from Harvard Medical School’s Center for Advanced Endoscopy delivered a state-of-the-art lecture on the advancements and future directions of Artificial Intelligence (AI) in endoscopy. Dr. Berzin is a consultant for multiple healthcare AI companies and his agenda covered various AI applications such as lesion measurement, computer-aided detection (CADe), computer-aided diagnosis, intra-operative navigation, exposure/prep assessment, semi-automated reporting, and multimodal generalist medical AI (GMAI).<br /><br />The lecture began with demystifying AI terminologies, the differences between traditional programming and machine learning, and addressing myths about AI. Examples included how machine learning simplifies predictive tasks and deep learning's capability of identifying features autonomously. Key distinctions between AI and human intelligence were highlighted, emphasizing AI's simultaneous power and fragility. He noted that current AI in medicine does not learn continuously, unlike consumer AI, due to FDA regulations.<br /><br />Debunking myths, Dr. Berzin clarified that AI robotic colonoscopy isn't imminent due to sensorimotor challenges that AI faces, reinforcing Moravec’s paradox. Key points about AI-powered colonoscopy discussed included lesion measurement, diagnosis, intra-op navigation, semi-automated reporting, and integrated AI systems.<br /><br />Dr. Berzin detailed the benefits and challenges of CADe, specifically for GI endoscopy. Despite CADe's proven benefits in improving adenoma detection rates (ADR) and polyp detection rates (PDR), its adoption has been slow due to economic considerations, lack of quality mandates, and complexities in the existing system. Dr. Berzin also touched upon AI’s potential to evolve beyond point solutions to offer integrated and broader functionalities.<br /><br />Highlighting the future of AI in GI endoscopy, Dr. Berzin emphasized the development of Generalist Medical AI (GMAI) which can interact with multiple data modalities and serve as a co-pilot in procedures by providing context-aware decisions. The journey towards AI adoption in GI practices involves ensuring safety, proving economic returns, and fostering innovation in clinical practice.

Keywords

Artificial Intelligence

Endoscopy

Dr. Tyler Berzin

Harvard Medical School

CADe

Generalist Medical AI

Lesion Measurement

Intra-operative Navigation

Machine Learning

GI Endoscopy