
Professor of Radiation Oncology @APHP @univ_paris_cite | AI researcher @INSERM | Author
Paris
Joined June 2026
- Tweets85
- Following30
- Followers0
- Likes0
1/4 Does adding anlotinib to consolidation immunotherapy help after chemoradiotherapy for unresectable stage III NSCLC? R-ALPS enrolled 553 patients across two stages.
1
3/4 More toxicity, uncertain added benefit. A longer median alone cannot establish superiority; the confidence interval matters.
1
4/4 Primary source: doi.org/10.1001/jamaoncol.20…
1/4 Can AI improve treatment decisions? A retrospective study compared initial AI recommendations with final physician decisions in 461 virtual visits.
1
1
3/4 This measures judged decision quality, not improved patient outcomes. Prospective, independent evaluation is still needed.
1
4/4 Primary source: doi.org/10.7326/ANNALS-24-03…
I’m a doctor. I built a free app to turn prevention into daily habits.
Health Points: 20 evidence-based habits (WHO, ESC), meal photos, Nutri-Score. Local AI. No account. No ads. No data collected.
iPhone: apps.apple.com/app/id6828323…
Android: play.google.com/store/apps/d…
14
Can an AI conduct a diagnostic consultation as well as a physician?
AMIE was compared with 20 primary-care physicians across 159 simulated, text-based cases in a randomized, blinded crossover study. 1/3
1
3
Specialists rated AMIE more accurate diagnostically, with higher scores on most dialogue-quality measures.
But these were patient actors in controlled text conversations, not routine consultations with real patients. 2/3
1
3
Promising performance, not proof of patient benefit or autonomous safety. Prospective evaluation must test errors, workflow and harm in real care.
Nature, 2025: doi.org/10.1038/s41586-025-0…
3/3
1/3 Can a mammogram flag breast cancer 4–6 years before diagnosis? A Norwegian retrospective cohort tested a commercial detection AI in 116,495 women across three biennial screening rounds.
3
1
28
2/3 AI scores were higher in the breast that later developed cancer. For screening-detected cancers, AUC rose from 0.64 at the earliest round to 0.97 at the diagnostic round; for interval cancers, it rose from 0.66 to 0.78.
1
14
3/3 This is a risk signal, not proof of patient benefit: one algorithm, retrospective data and limited diversity. Prospective trials must test effects on advanced cancer, unnecessary imaging and mortality. jamanetwork.com/journals/jam…
3
1/3
An AI can analyze 39 features of an echocardiogram. How should we use it?
PanEcho was developed and validated using 1.2 million videos from 32,265 examinations in 24,405 patients, with additional external validation.
1
10
2/3
It performed 18 classification tasks with a median internal AUROC of 0.91, plus 21 measurement tasks.
For left ventricular ejection fraction, mean absolute error was 4.2 percentage points internally and 4.5 externally.
1
7
3/3
Potential value: an adjunct reader that helps standardize measurements.
Retrospective validation. Prospective testing must assess workflow benefit, important errors and poor-quality acquisitions.
doi.org/10.1001/jama.2025.87…
1/4
A doctor’s final approval can still follow an AI-induced error.
In a randomized experiment, 225 physicians assessed neurocritical-care cases with the same AI instructed to support correct decisions or covertly steer them toward harmful ones.
1
4
3/4
This does NOT estimate the harm rate of ordinary clinical AI: the adversarial model was deliberately instructed to mislead.
It tests whether medical expertise alone resists that manipulation. Experience offered no clear protection in this experiment.
1
1
4/4
Evaluate how AI advice changes clinical decisions, including initially correct ones. A final human signature is an incomplete safety measure.
Preprint, not peer-reviewed:
doi.org/10.64898/2026.10.05.…


