KonstantinosBalaskas

@rohanpaul_ai Spot-on—your MedAgentAudit breakdown shows how accuracy hides "broken teamwork" in med AI. HERMES cRCT echoes: DeepMind AI AUC 0.99 → 0.56 real-world, 2x false referrals vs. clinicians. Teleophthalmology cuts false-positives 59%. Evidence > illusions. Lancet: x.com/KonBalaskas/st… (DM for NIHR economic eval)

This is not a great news for Medical AI. ☹️ Shows that medical AI Agents built from multiple LLMs often look correct but think wrong. Even when they give the right diagnosis, most of them got there through broken teamwork. In over 68% of “successful” cases, the agents already agreed at the start, so the group talk did nothing useful. The authors reviewed 3,600 medical cases and found repeating problems, like correct facts disappearing mid-discussion, smart minority opinions getting ignored, and agents choosing easy votes instead of reasoning. They built a tool called AuditTrail to track how facts and opinions move during a debate. It found that evidence often drops between early steps and the final answer, and that longer talks sometimes make things worse because the models start copying each other instead of thinking. Even worse, the systems often pick low-risk answers when a life-threatening one was on the table. So the scary part is this, a “high-accuracy” medical AI can sound smart but still hide unsafe logic underneath. Overall this paper exposes how accuracy numbers lie. An AI can pass medical tests yet still reason in unsafe, careless, or biased ways, which makes it untrustworthy for real patients. --- Paper – arxiv. org/abs/2510.10185 Paper Title: "MedAgentAudit: Diagnosing and Quantifying Collaborative Failure Modes in Medical Multi-Agent Systems"

An amazing team effort 🤝 - Big thank you 🙏 to all our collaborators, community optometrists, hospital teams, and our wonderful study participants for making this possible! 👏 #DigitalTransformation @Moorfields #NIHR

HERMES: Beyond the main trial🫶 ✅ Human-Computer Interaction: Inspiring work by the brilliant Ann Blandford and team @uclic 🔐 to real-life clinical deployment & digital inclusion 👉Links to papers doi.org/10.1177/205520… doi.org/10.1136/bmjope…

🚀 New paper in The Lancet Primary Care! Proud to share the results of our NIHR HERMES study 💥 👉 Links to paper & podcast: 📄 doi.org/10.1016/j.lanp… 📻 Podcast: In conversation with... Konstantinos Balaskas and Anitta Sharma thelancet.com/doi/story/10.1…

Excellent morning @BBCNews @BBCSport studios Thanks for having me again @MediaCityUK #manchester

Grateful to present data related to Natural history and genotype-phenotype correlation of patients with IQCB1 retinopathy (#NPHP5retinopathy) with or without #SeniorLokensyndrome at #ARVO2024 @ARVOinfo linkedin.com/posts/sagnik-s…

Our preprint researchsquare.com/article/rs-211…

@KonBalaskas @DunJackFu @Ismail_Moghul Amazing work! 16 papers/presentation at #ARVO2024 for #BalaskasLab!

Compelling evidence for the role of Photoreceptor Degeneration as the key predictor for rate of growth in #GeograhicAtrophy👇

Pallavi Bagga Gunjan Naik @GongyuZhang Ana Paula Ribeiro Reis Estelle Ioannidou Rosana Lima Georgina Wignall 👏❤️

Hot off the press 👇‼️ jamanetwork.com/journals/jamao… Very proud to see the first AI-enabled analysis of Geographic Atrophy OCT feature changes over 24 months from the phase III DERBY & OAKS studies published by our team in @JAMAOphth 😀@MoorfieldsBRC @UCLeye 🙌