Robin Demesmaeker, PhD

A news that I wanted to share: ChatGPT just launched a medical version And yeah, it’s impressive. This new mode is designed to help patients better understand their lab results, check symptoms, and even prep smarter questions before seeing a doctor. It’s fast & confident : shows PhD-level reasoning, tops benchmarks like HealthBench, and handles huge complexity in seconds. Honestly? It’s kind of wild how far it’s come 🤯 But at the same time, I’ve got questions. Because let’s be real: even with regular ChatGPT, a lot of people were already asking it for health advice, skipping the doctor, and self-prescribing based on what it said. Now that there’s a medical version? It’s going to feel even more official. Even more like : « I don’t need a doctor anymore. » But here’s the thing : it’s still not a doctor. 🧪 It gets about 55% accuracy on medical queries 💬 And it misses half of the way people actually describe symptoms : the everyday language we all use when something feels off And beyond all that, health isn’t just about facts or data. It’s about context. Emotion. Timing. Human instinct. Connection. So yeah, GPT-5 in medicine is a big step forward. But also one that comes with risk. Especially if people start trusting it too much. 👉 Curious to see how this plays out in real-world care

127

35 Kommentare

AI now diagnoses at 92% accuracy—outperforming physicians at 74% in randomized trials. Yet adding GPT-4 to clinicians doesn’t yield consistent improvement in diagnosis or treatment. Why? New Stanford AI Index 2025 data reveals a paradox: LLMs are impressive at clinical inference, but real-world workflows haven’t caught up. Some highlights: 🩺 Diagnosis & Treatment – GPT-4: 92% diagnostic accuracy vs. physicians at 74% – Treatment decisions: 84.7% vs. 78.2% (GPT-4 vs. MDs) – No clear synergy when humans + GPT-4 combine → signals deeper issues than “just integrate it” 📈 Benchmarks Hitting a Ceiling – MedQA at 96% accuracy (zero fine-tuning) – Static question-answer sets do not reflect the messiness of real clinical practice – Next frontier: benchmarks that reflect real-world complexity and nuance 💰 Cost vs. Outcomes – OpenAI's “o1” sets new records on 19 clinical tasks but is ~1.5× more expensive than GPT-4 – Raises questions on ROI for large-scale hospital deployment 🤖 Ambient AI Scribes – ~20 minutes less EHR work/day – Burnout down 26%, admin burden down 35% – Next step: from “listen + transcribe” to “listen + act” 🔬 Research & Trials – 1,200+ healthcare LLM papers in 2024, doubling yearly – 537 AI-specific clinical trials; U.S. & China lead globally – FDA: 223 AI med devices approved in 2023 vs. single digits a decade ago Full 18-page chapter for deeper data: team spent months on this. Great working on this one with co authors Stephen Ma, Jonathan H. Chen and other researchers + faculty. AI is surpassing physicians on standardized tasks—but how might that reshape the physician’s role? Are we ready—legally, ethically, and operationally—for a future where AI drives diagnosis and treatment plans, with physicians offering higher-level oversight and patient advocacy? If data-driven tasks become largely automated, which human skills gain new importance, and how should medical education adapt? #StanfordHAI #AIIndex #AIinHealthcare #DigitalHealth #HealthcareInnovation #HealthTech #ClinicalAI

444

69 Kommentare

I Believe It’s Time to Rethink Scientific Storytelling   I’ve been in science communication long enough to know one thing for sure: Good data doesn’t speak for itself. It needs a voice. And most of our scientific content? It sounds like a robot whispering into a void.   A few months ago, we ran a post-campaign insight study. Same data. Same product.   Two versions of communication: 1. Traditional MoA deck 2. Narrative-led version with real-world context and a patient-centered angle   Combined Outcome: - 2.5× HCP engagement - 43% higher message recall - Deeper rep–doctor conversations   It hit me again: The science hasn’t changed. The way we tell it has to.   I say this as someone who’s worked with pharma, medtech, diagnostics, and clinical-stage biotechs: We’ve over-indexed on accuracy… and underinvested in emotion, clarity, and connection. This isn’t about oversimplifying. It’s about translating.   🔹 Mechanism → Meaning 🔹 Molecules → Human moments 🔹 Data → Decisions   Disclaimer: Based on Healthminds Consulting experience   #healthminds #healthcareprofessionals #pharma #medical #lifescience #clientresults #outcomes #scientificstorytelling #sciencecommunication #pharmaceuticalindustry

97

10 Kommentare

Open-source AI is reshaping healthcare. Here’s a curated list of clinical, imaging, and biomedical models. Full details and links are available in the Notion guide below. 👇 📌 Clinical & Biomedical Text Models • BioGPT – Generative biomedical Transformer • Clinical-T5 – Text-to-text clinical modeling • PubMedBERT – Pretraining on biomedical abstracts • ClinicalBERT – Clinical notes QA, NER • BioELECTRA – Biomedical text encoding • BioMegatron – Large-scale biomedical representation • BlueBERT – Clinical and biomedical text understanding • Med-BERT – EHR-based disease prediction • BioMedLM – Biomedical GPT model for research use • KeBioLM – Biomedical NER and relation extraction • LinkBERT – Linked pretraining across documents • BioBART – Biomedical dialogue and summarization • PMC-LLaMA – Biomedical LLaMA finetuned on scientific articles • ChatDoctor – Healthcare chatbot from LLaMA • CPLLM – Clinical prediction using diagnosis sequences • LLaMA-2 – General dialogue and clinical information retrieval • Med-PaLM2 – Medical licensing exam QA • HuatuoGPT-II – Medical consultation modeling • DoctorGLM – Clinical question answering trained on CMD data 📌 Medical Imaging AI Models • MONAI Label – Open-source medical imaging annotation framework • ChatCAD – Automated generation of imaging reports • FastFold – Accelerated AlphaFold for protein folding • SwinIR-Med – Medical image super-resolution and restoration • DragonFlyMed – Biomedical visual reasoning and fine-grained analysis • RadBERT – NLP on radiology reports • MAIRA-2 – CXR radiology report generation • VINDR-CXR – Open models and datasets for chest X-ray diagnosis • Merlin – 3D abdominal CT interpretation with few-shot learning • MedSAM2 – Medical adaptation of the Segment Anything Model (SAM) 📌 Multimodal Clinical Decision Models • BioViL – Vision-language model for radiology understanding • Meta Fair – Unified text-image multimodal generation • Meditron – Clinical diagnostics powered by open LLaMA models • Med-Flamingo – Visual and textual clinical reasoning _________________________________________________ #llm #ai #medicine #healthcare #machinelearning

1.654

74 Kommentare

🚀 Exciting News from iSimulate & 3B Scientific! We’re thrilled to announce the REALITi Scenario Converter GPT — the fastest way to create high-fidelity, clinical-grade simulation scenarios for the REALITi platform. ✅ Upload clinical cases (text, document, spreadsheet, image) ✅ Auto-generate .isim simulation files ready for upload ✅ Built-in clinical validation for vitals, labs, and ECG rhythms ✅ Dynamic placeholder handling (patient name, age, condition) ✅ New Scenario Assist feature for smart clinical recommendations ✅ Supports full ABCDE structure, AVPU-to-GCS mapping, panels, and more! 🔗 Access it here: https://lnkd.in/g8NpAktM ⚡ Requires a free OpenAI ChatGPT account to use the GPT. Developed by iSimulate in collaboration with 3B Scientific — bringing innovation to healthcare simulation, education, and clinical training globally. 🌍 #HealthcareSimulation #REALITi #SimulationEducation #MedicalEducation #iSimulate #3BScientific #Innovation #ParamedicTraining #SimulationBasedEducation

28

I've done something different this week: There is always a lot of AI News. And this post aims to make that easier for you. So this week I have filtered to the two most important bits of news in each section. Yay or nay? 𝐔𝐊 𝐇𝐞𝐚𝐥𝐭𝐡𝐜𝐚𝐫𝐞 𝐀𝐈 𝐍𝐞𝐰𝐬: 🇬🇧 The Life Science Sector Plan was published with a few notable quotes on AI like the impact of bespoke IT on deployment (page 20), how AI will impact the sector (page 24), a new MHRA framework for AI in 2026 (page 34). https://lnkd.in/eX8fD-FZ 🇬🇧 NHS 111 Wales has launched an AI assistant to help you navigate their website. https://lnkd.in/e3_Zs_SG  𝐎𝐭𝐡𝐞𝐫 𝐫𝐞𝐥𝐚𝐭𝐞𝐝 𝐔𝐊 𝐍𝐞𝐰𝐬: 🇬🇧 The Government is seeking Responsible AI Advisory Panel members. https://lnkd.in/ePW9Miub 🇬🇧 Tech entrepreneurs outside of London will be able to get funding and skills development to support their growth. https://lnkd.in/e336HgjX 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐀𝐈 𝐧𝐞𝐰𝐬: 🇺🇸 In the latest edition of the US AI dynamic, the California State Senator introduced a Bill amendment that would require big tech companies to publish on safety and security. https://lnkd.in/eaCjkhmH 🇪🇺 The co-chairs of the European Parliament’s working group has given the GPAI finalised code of practice the nod, and the code has already been signed by OpenAI. https://lnkd.in/eqWyQ37z  https://lnkd.in/eXWtVZyy 

22

9 Kommentare

AI for Colonic Polyp Detection: One Dataset, Three Guidelines, Diverging Recommendations. The same data, three major guidelines — and yet different takes on how AI should be implemented in clinical practice. 💡 Want to know whether AI tools are being recommended — and where the disagreements lie? We've got the info. 🎥 Watch now to see how the future of endoscopic AI is being shaped https://lnkd.in/dX2cSngg, or #AIinHealthcare #Colonoscopy #MedicalAI #ClinicalGuidelines #PolypDetection #Endoscopy #Gastroenterology #HealthTech

31

🔥 Wow....this is big! FDA just launched Elsa, a generative AI tool designed to help their staff to do scientific review and investigation more efficiently 🌐 See link for further details: https://lnkd.in/gJpTvDeB 👨‍💻 Elsa in Action, according to the media release: 💊 Accelerate clinical protocol reviews 💊 Shorten the time needed for scientific evaluations 💊Identify high-priority inspections targets 💊 Perform faster label comparisons 💊 Summarize adverse events to support safety profile assessments 💊 Generate code to help develop databases for nonclinical applications 📊 This completes the "Approval" portion of AI application in drug dev pipeline (in the diagram below) -see link for article. ✨ Next question will be - not if, but when can IRB/EC starts using AI to assist with the review? ✨How will participants and community be informed? ✨and Quality of the review? - What do you think? 👍Like, 🔄Share ⌨Comment and 🔖Tag a friend #artificialintelligence #ai #drugdiscovery #drugdevelopment #clinicaltrials #lifesciences #biotech #pharmaceutical #researchvalet #clinicalresearch #innovation #healthcare #medicalresearch

174

41 Kommentare

#AI is not new, #AI is not magic, #AI is not intelligent," according to Jessica Rose Morley, PhD, who at #HIMSS25Europe delivered a fantastic talk on the myths and realities of #AI in #healthcare. Here are my notes from the speech: ✅ $450B by 2030 - the projected global AI healthcare market. Dr. Morley warns against viewing this investment as a magic fix. “Most policy documents present AI as if it can solve everything: improve care, cut costs, enhance equity. But these goals often directly contradict each other. ✅ AI is not new. What’s changed is the scale. AI now works at a systems level, reshaping how care is delivered. ✅ AI treats data twins, not people. The shift to algorithm-based medicine means we're moving from patient-centric to data-centric care. The patient becomes their “data shadow”, an imperfect proxy that rarely captures the whole person. ✅ 40%. That's the upper limit of how much individual health outcomes can be predicted from personal data. The rest is shaped by social and environmental factors, areas where current AI models fall short. ✅ From narrow trust to distributed trust. AI breaks the old model of accountability in medicine. When something goes wrong, is it the algorithm, the data, or the doctor who’s at fault? In a system of uncertainty, system-level trust becomes essential. ✅ Inverse Data Quality Law: The availability of high-quality health data tends to vary inversely with the need of the population served - the people who need care most are often the least visible to AI. ✅ Her recommendation: shift funding from precision medicine to population health, invest in social determinants of health, and build digital health literacy for everyone. ✅ To ensure equity and effectiveness, AI must be implemented with humility, transparency, and a focus on people, not just their data.

44

4 Kommentare

BREAKING NEWS! Two popular large language models (LLMs), GPT-4 and Llama-3, produced medical device-like decision support across a range of scenarios even though those are not regulated by the FDA. Authors of a new study were interested in finding out whether LLMs show promise for clinical decision support (CDS) even though none is currently authorized by the FDA as a CDS device.  "𝐹𝑜𝑙𝑙𝑜𝑤𝑖𝑛𝑔 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑠𝑢𝑝𝑝𝑜𝑟𝑡 𝑟𝑒𝑞𝑢𝑒𝑠𝑡𝑠 𝑎𝑏𝑜𝑢𝑡 𝑡𝑖𝑚𝑒-𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 𝑒𝑚𝑒𝑟𝑔𝑒𝑛𝑐𝑖𝑒𝑠, 𝟣𝟢𝟢% 𝑜𝑓 𝐺𝑃𝑇-𝟦 𝑎𝑛𝑑 𝟧𝟤% 𝑜𝑓 𝐿𝑙𝑎𝑚𝑎-𝟥 𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒𝑠 𝑤𝑒𝑟𝑒 𝑐𝑜𝑛𝑠𝑖𝑠𝑡𝑒𝑛𝑡 𝑤𝑖𝑡ℎ 𝑑𝑒𝑣𝑖𝑐𝑒-𝑙𝑖𝑘𝑒 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑠𝑢𝑝𝑝𝑜𝑟𝑡 (𝐹𝑖𝑔. 𝟣). 𝐴𝑙𝑙 𝑚𝑜𝑑𝑒𝑙 𝑠𝑢𝑔𝑔𝑒𝑠𝑡𝑖𝑜𝑛𝑠 𝑤𝑒𝑟𝑒 𝑐𝑙𝑖𝑛𝑖𝑐𝑎𝑙𝑙𝑦 𝑎𝑝𝑝𝑟𝑜𝑝𝑟𝑖𝑎𝑡𝑒 𝑎𝑛𝑑 𝑐𝑜𝑛𝑠𝑖𝑠𝑡𝑒𝑛𝑡 𝑤𝑖𝑡ℎ 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑𝑠 𝑜𝑓 𝑐𝑎𝑟𝑒." Conclusions? 1) LLMs are capable of CDS! 2) If so, people will use them on a massive scale. 3) If so, it would make a lot of sense for regulators to provide regulatory oversight even though LLMs provide disclaimers that they are not intended to be used for medical advice. And these are just unimodal LLMs. Imagine the regulatory challenge that will come with multi-modality. Study: https://lnkd.in/eaCCbXed

66

12 Kommentare