NEJM AI Grand Rounds Pixels to Protocols: Building the AI Future of Pathology with Dr. Faisal Mahmood
9 snips
Apr 16, 2025 Dr. Faisal Mahmood, an Associate Professor of Pathology at Brigham and Women’s Hospital, shares his groundbreaking work in AI and pathology. He discusses how foundational models can transform digital diagnostics and reflects on the power of open-source culture for innovation. Faisal explores the development of PathChat, a generative AI co-pilot, and its FDA breakthrough designation. He also ponders the future of patient-level models and the potential of AI to revolutionize healthcare through real-time data analysis.
AI Snips
Chapters
Books
Transcript
Episode notes
CLAM for Cancer Primary Identification
- CLAM addressed weakly supervised learning with large pathology slide images, using attention and pretrained features for data efficiency.
- It was motivated by finding primary origins of cancers with unknown primary, a challenging clinical task.
Publish Code to Accelerate Progress
- Make code and models publicly available to accelerate community feedback and improvements.
- Open sourcing enables unexpected uses and helps refine pipelines effectively.
From Foundations to Usable AI
- Transform strong foundation models into usable clinical tools with instruction fine-tuning and reinforcement learning from human feedback.
- This process makes AI models more responsive and effective in real-world medical use cases.
