600+ algorithms cleared by FDA
Huge impact of multimodal for multiple clinical applications.
The talk with be on youtube.
The main paper cited during the talk:
- Topol EJ. Medical forecasting. Science. 2024 May 24;384(6698):eadp7977. doi: 10.1126/science.adp7977. Epub 2024 May 23. PMID: 38781357.
- Moor M, Banerjee O, Abad ZSH, Krumholz HM, Leskovec J, Topol EJ, Rajpurkar P. Foundation models for generalist medical artificial intelligence. Nature. 2023 Apr;616(7956):259-265. doi: 10.1038/s41586-023-05881-4. Epub 2023 Apr 12. PMID: 37045921.
Medical forecasting
- Define risk: Partitionning high risk patients
- Lifestylefactors : UPF in diet, exercise, air pollution, sleep
- Genetics : APOE
- Biomarkers : p-tau 217
- PRS
- Retina photo
- You can predict the future of alzheimer and parkinson disease years before with the retina photo
- Monitoring high risk people
- Assess them more frequently :
- check retina
- brain organ clocks.: Tony Risckerey (?), validated in UK biobank
- epigenetic
- biomarkers : p-tau 217
- Timing of intervention: when will it begin to manifest and develop?
- La place du multimodal AI
- Preventive therapy
- Anti-inflammatory : e.g. GLP1
- Anti-amyloid
- no simple drug
- we will hopefully have more drugs and more efficient → but we need step 1 2 and 3 to build accurate clinical trials!
- they also will help to assess the ruled-out drugs with greatest subgroup analysis
- Full court press Lifestyle factors
Challenges of multimodal AI
- inputs data : text/audio/images/sensors/genomics
- context = medical domain knozledge : the entire corpus of medical literature + knowledge graph
- outputs and applications:
- virtual health coach: highly exciting
Medical education
- Anyone working in healthcare needs to be educated about AI
- embodied biases
- when to trust it or not
- etc
Health inequity and biases