Machine Learning & Healthcare
summary
Claims that algorithms outscore radiologists hinge on narrow setups not real work: CheXNet vs radiologists uses single frontal x‑rays with no prior studies or clinical context; published papers often cherry‑pick high‑confidence subsets (e.g. 30,000 CT studies ≈5M images but reporting 0.037% FN on top 8.5% of cases). Key limitations are weak NLP labels, interobserver variability, artifacts and low positive prevalence. Practical wins come from workflow systems and targeted models like spinal cord compression: 70,000 MRIs and a DeepSpine project with prospective validation and EHR integration.
AI-written summary · about
chapters (7)
- 0:00AI in healthcare and radiology
- 1:52Hype around AI in radiology
- 2:12Reality check on AI radiology
- 7:48Why medical imaging is hard
- 16:30AI beyond image recognition
- 18:14Future of radiology with AI
- 22:52Q&A
related talks
36mAI in Healthcare
37mAI in medicine: where do we stand?
47mAI for making decisions in clinical medicine
09mHow AI is changing the healthcare system
43mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
25mDeep Learning in Biomedicine: Challenges and Opportunities
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
28mThe Future of Healthcare – AI’s Promise and the Challenges Ahead