AI for making decisions in clinical medicine
summary
Describes building self‑supervised models that pretrain on unlabeled medical images and audio then fine‑tune on small labeled sets, demonstrates zero-shot classification by matching x‑rays to reports, and ships datasets and benchmarks like CheXphoto competition and CheXpert. Reports concrete results: assisted clinicians rise from 0.60 to 0.65 accuracy while the standalone model reaches 0.79, and one radiology use case avoids five unnecessary ACL surgeries per 100 negatives. Forms the Medical AI Validation Consortium spanning 16 cities and 60+ institutions to test generalizability.
AI-written summary · about
chapters (6)
- 0:00Introduction to medical AI
- 5:45Developing label-efficient algorithms
- 14:21Clinician-AI collaboration studies
- 22:03Open benchmark curation challenges
- 31:38Global validation consortium
- 33:24Community resources and Q&A
related talks
37mAI in medicine: where do we stand?
23mAI in the Life Science Sector
36mAI in Healthcare
25mAI for healthcare: Scaling Access and Quality of Care for Everyone
36mResponsible AI
44mAI in the Clinic: Prediction is Better Than Cure
09mHow AI is changing the healthcare system
28mMachine Learning & Healthcare