<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Tech Anatomy · News</title><link>https://techanatomy.co/news.html</link><atom:link href="https://techanatomy.co/news.xml" rel="self" type="application/rss+xml"/><description>Analysis on AI and health technology for health and pharma leaders.</description><language>en</language><lastBuildDate>Sun, 08 Nov 2026 08:00:00 +0000</lastBuildDate><item><title>Hospitals have AI products. Why so little of it is in daily use, and what the usual numbers do and do not show</title><link>https://techanatomy.co/a/ai-adoption-last-mile-hospitals.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-adoption-last-mile-hospitals.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Commercial AI tools for radiology have existed for years, yet a 2020 survey found only about 30% of responding radiologists using AI in practice. Jan Beger, Global Head of AI Advocacy at GE HealthCare, points to strategy and AI literacy rather than technology, and to a set of figures that are widely repeated in the case for AI.</description></item><item><title>What AI agents are, and what has changed since the first ones appeared</title><link>https://techanatomy.co/a/ai-agents-explained-healthcare.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-agents-explained-healthcare.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>An AI agent is a language model that can plan and act, not only answer. When Tech Anatomy explained the idea in January 2025, the first agents were research previews; by 2026 they were built into mainstream products, including offers aimed at healthcare, while the evidence on what they deliver in practice remains thin.</description></item><item><title>Eight years of AI for Health: the barriers that keep returning to the agenda</title><link>https://techanatomy.co/a/ai-in-health-recurring-barriers-data-access.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-in-health-recurring-barriers-data-access.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>After eight editions of the AI for Health summit in Paris, its director Sébastien Marguerès says the same subject comes back every year: access to health data, followed by data volume and hospital infrastructure. The European and French responses are real, and most of them have not yet taken full effect.</description></item><item><title>Ten years of the Cercle P: how a pharma-run think tank tries to turn patient testimony into policy</title><link>https://techanatomy.co/a/cercle-p-patient-testimony-to-policy.html</link><guid isPermaLink="true">https://techanatomy.co/a/cercle-p-patient-testimony-to-policy.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Pfizer France created the Cercle P in 2016 to think about the place of patients in tomorrow&#x27;s medicine. Ianis Ménot, who runs it, describes its method and two proposals that later appeared in French regulation, which raises the question of how to judge the influence of such a group.</description></item><item><title>Chain-of-thought, the prompting technique that taught language models to show their work</title><link>https://techanatomy.co/a/chain-of-thought-reasoning-llms.html</link><guid isPermaLink="true">https://techanatomy.co/a/chain-of-thought-reasoning-llms.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>In 2022 Google researchers showed that asking a large language model to write out intermediate steps roughly tripled its score on school maths problems. The idea now underpins the &quot;reasoning&quot; models used in medicine, and later research has complicated the hope that the written steps reveal how the model actually reached its answer.</description></item><item><title>Twenty-five years of digital health trials, mapped by strategy</title><link>https://techanatomy.co/a/digital-health-interventions-25-years.html</link><guid isPermaLink="true">https://techanatomy.co/a/digital-health-interventions-25-years.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>A 2025 review in the Journal of Medical Internet Research catalogued the techniques used in 885 randomised trials of digital health interventions published since 1999. It shows what designers have relied on, and where the research has concentrated, more than what works.</description></item><item><title>Digital twins in health, from a metaverse avatar to a model of one heart</title><link>https://techanatomy.co/a/digital-twins-health-avatar.html</link><guid isPermaLink="true">https://techanatomy.co/a/digital-twins-health-avatar.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>The idea of a personal health avatar fed by wearables and medical records has circulated since the metaverse boom of the early 2020s. The digital twins that have reached clinical research are narrower, built around one organ and one decision, and the question of who controls the data behind them is still open.</description></item><item><title>How a heartbeat on a smartwatch becomes an AI prediction</title><link>https://techanatomy.co/a/from-wearable-signal-to-ai-model.html</link><guid isPermaLink="true">https://techanatomy.co/a/from-wearable-signal-to-ai-model.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Between the green light on the back of a watch and an alert on a phone, a heart rate measurement goes through several transformations, some on the device and some on remote servers. Following one data point along that path shows where the processing happens, what kinds of models are involved and which protections apply.</description></item><item><title>A health digital twin for prevention: what Integratome models, and where decision support ends</title><link>https://techanatomy.co/a/health-digital-twin-prevention-integratome.html</link><guid isPermaLink="true">https://techanatomy.co/a/health-digital-twin-prevention-integratome.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Integratome, a French start-up, maps a person&#x27;s questionnaire and lab results onto a probabilistic model of around sixty chronic conditions and projects their risks five years ahead. Its CEO Ghislain de Kerviler describes it as augmented medicine for doctors, not a medical device, a position that European rules on prediction software make worth examining.</description></item><item><title>Where bias enters a healthcare AI model, and what can be done at each stage</title><link>https://techanatomy.co/a/hidden-bias-healthcare-ai.html</link><guid isPermaLink="true">https://techanatomy.co/a/hidden-bias-healthcare-ai.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>A 2025 review in npj Digital Medicine traces how bias gets into medical AI, from the data a model learns from to the way clinicians use its output. The best-documented cases show that the problem often lies in what a model is asked to predict, not in the algorithm itself.</description></item><item><title>Seventy years of AI in medicine, from expert systems to language models</title><link>https://techanatomy.co/a/history-of-ai-in-medicine.html</link><guid isPermaLink="true">https://techanatomy.co/a/history-of-ai-in-medicine.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Medical AI did not start with ChatGPT. Rule-based systems were diagnosing infections and internal diseases in the 1970s, and several matched specialists in evaluations without ever entering routine care. Each wave since has faced a version of the same gap between performance in a test and use in a clinic.</description></item><item><title>Machine learning in five steps, and where each one can go wrong in healthcare</title><link>https://techanatomy.co/a/machine-learning-five-steps-in-healthcare.html</link><guid isPermaLink="true">https://techanatomy.co/a/machine-learning-five-steps-in-healthcare.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Most clinical machine learning follows the same sequence: collect labelled data, clean it, choose a model, train it, test it. Three classic hospital tasks show how the sequence works, and the published record shows which step tends to decide whether a model helps.</description></item><item><title>Building a medical assistant on guidelines rather than the open web: the design choices behind MedGPT</title><link>https://techanatomy.co/a/medgpt-guideline-based-medical-assistant.html</link><guid isPermaLink="true">https://techanatomy.co/a/medgpt-guideline-based-medical-assistant.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Synapse Medicine launched MedGPT, a conversational assistant for French health professionals, in September 2025. Its co-founder Louis Létinier explains how it is built, why it answers from national guidelines first, and how the company intends to fund it.</description></item><item><title>Should patients help build health AI? Nine proposals from the Cercle P and the questions they leave open</title><link>https://techanatomy.co/a/patients-in-health-ai-cercle-p-proposals.html</link><guid isPermaLink="true">https://techanatomy.co/a/patients-in-health-ai-cercle-p-proposals.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>In a small survey run by the Cercle P, seven in ten patients said they had never used AI in their care, yet most wanted to be told whenever it is used. Three members of the group explain the nine proposals that followed, on information, participation and public policy.</description></item><item><title>Quantum computing for medicine, explained without the hype</title><link>https://techanatomy.co/a/quantum-computing-medicine-explained.html</link><guid isPermaLink="true">https://techanatomy.co/a/quantum-computing-medicine-explained.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Quantum computers work on qubits rather than bits, and their most credible medical use is the simulation of molecules. The hardware has improved quickly since 2024, but no medical problem has yet been solved on one that a conventional computer could not handle.</description></item><item><title>Synthetic health data: how it is made, where it is used, and what it does not guarantee</title><link>https://techanatomy.co/a/synthetic-health-data-uses-and-limits.html</link><guid isPermaLink="true">https://techanatomy.co/a/synthetic-health-data-uses-and-limits.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Synthetic data is artificial patient data generated by algorithms to mimic the patterns of real records. It is already used to share data, enrich rare cases and design trials, while researchers still debate how much privacy it actually protects.</description></item><item><title>Why most of the data in a medical record is never reused</title><link>https://techanatomy.co/a/why-health-data-goes-unused.html</link><guid isPermaLink="true">https://techanatomy.co/a/why-health-data-goes-unused.html</guid><pubDate>Sun, 08 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Electronic health records hold notes, prescriptions, results and images, yet little of it is analysed beyond the consultation that produced it. Privacy law, fragmented systems and unclear incentives all play a part, and synthetic data and digital twins are two of the responses now being tested.</description></item><item><title>Sixty years of computers in drug discovery, and what has reached the clinic so far</title><link>https://techanatomy.co/a/ai-drug-discovery-history.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-drug-discovery-history.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Chemists began predicting what a molecule would do before making it with an equation on paper in 1964. The first candidates designed with AI have now been through human trials, and the question has shifted to whether they succeed more often than other drugs.</description></item><item><title>The evidence dilemma: how a surgeon turned legislator thinks about regulating AI in health</title><link>https://techanatomy.co/a/ai-evidence-dilemma-health-policy.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-evidence-dilemma-health-policy.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>A UN scientific panel warns that the evidence needed to govern AI arrives after the decisions have to be made. Andrei Baciu, a cardiovascular surgeon in the Romanian Parliament, describes how health systems and lawmakers might act in the meantime.</description></item><item><title>AI in cardiology: what the ECG studies show, and what they leave open</title><link>https://techanatomy.co/a/ai-in-cardiology-ecg-evidence.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-in-cardiology-ecg-evidence.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Algorithms trained on standard ECGs can flag a weak heart pump or a likely history of atrial fibrillation that a cardiologist cannot see on the tracing. Electrophysiologist Benedict Glover, who has also built an AI tool of his own, explains where this helps in practice and why the outcome data are still missing.</description></item><item><title>AI in medical imaging, ten years after &quot;stop training radiologists&quot;</title><link>https://techanatomy.co/a/ai-medical-imaging-story.html</link><guid isPermaLink="true">https://techanatomy.co/a/ai-medical-imaging-story.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>In 2016 Geoffrey Hinton predicted that deep learning would outperform radiologists within five years. The FDA has since authorised more than 1,600 AI-enabled devices, most of them in imaging, and radiology departments are still hiring. What happened in between says a lot about the distance between a benchmark and a clinic.</description></item><item><title>What AlphaFold solved, and what it left to the rest of drug discovery</title><link>https://techanatomy.co/a/alphafold-protein-structure-story.html</link><guid isPermaLink="true">https://techanatomy.co/a/alphafold-protein-structure-story.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>In 2020 a DeepMind system predicted protein shapes with accuracy close to laboratory methods, and in 2024 its designers shared a Nobel Prize. The prediction problem is largely answered. What it changes for medicine is a separate question.</description></item><item><title>Second Sight and the Argus II: what happens to an implant when its maker moves on</title><link>https://techanatomy.co/a/argus-ii-bionic-eye-orphan-implants.html</link><guid isPermaLink="true">https://techanatomy.co/a/argus-ii-bionic-eye-orphan-implants.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>The Argus II was the first retinal implant approved in the United States, and more than 350 blind people received it. When Second Sight stopped making it, the device stayed in their eyes while the support around it disappeared.</description></item><item><title>Bioprinting after Organovo: what printed tissue does today, and what still stands between it and an organ</title><link>https://techanatomy.co/a/bioprinting-organovo-and-vascularisation.html</link><guid isPermaLink="true">https://techanatomy.co/a/bioprinting-organovo-and-vascularisation.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Bioprinted human liver tissue has been sold for drug testing since 2014, yet no printed vital organ has been transplanted into a patient. Most of the distance comes down to one biological problem: supplying blood to the inside of thick tissue.</description></item><item><title>Consumer DNA tests after 23andMe: what a saliva kit reads, and who keeps the data</title><link>https://techanatomy.co/a/consumer-dna-tests-23andme.html</link><guid isPermaLink="true">https://techanatomy.co/a/consumer-dna-tests-23andme.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>23andMe turned genetic testing into a consumer product, won FDA authorisations, lost data from 6.9 million people and went through bankruptcy. The story raises two separate questions: what these tests can actually tell a customer, and who should hold the data they produce.</description></item><item><title>Decentralized science five years on: what DAOs fund, and what they have not yet shown</title><link>https://techanatomy.co/a/desci-decentralized-science-funding.html</link><guid isPermaLink="true">https://techanatomy.co/a/desci-decentralized-science-funding.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Since 2021, communities organised on blockchains have pooled money to fund early research in longevity and women&#x27;s health, and a Pfizer fund joined one of them. The amounts remain small next to public funders, and the decisive test, a therapy that reaches patients, is still years away.</description></item><item><title>Digital endpoints in clinical trials: what it takes to turn a sensor signal into evidence</title><link>https://techanatomy.co/a/digital-endpoints-clinical-trials-validation.html</link><guid isPermaLink="true">https://techanatomy.co/a/digital-endpoints-clinical-trials-validation.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>In 2023, the European Medicines Agency accepted a wearable-derived measure as a primary endpoint for Duchenne muscular dystrophy trials, about fourteen years after work on it began. Sumu Sethi, senior director for digital health at GSK, describes the validation path and the questions sponsors ask before adding a sensor to a trial.</description></item><item><title>After CE marking: what a digital medical device still has to prove in France</title><link>https://techanatomy.co/a/digital-medical-devices-after-ce-marking.html</link><guid isPermaLink="true">https://techanatomy.co/a/digital-medical-devices-after-ce-marking.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>In a three-part Tech Anatomy series, Snitem president Laurence Comte-Arassus and Tilak Healthcare chief executive Édouard Gasser describe the distance between regulatory approval and real use of a digital medical device. Tilak&#x27;s own path, from Article 51 to a negative HAS opinion and then early coverage under PECAN, runs through the whole conversation.</description></item><item><title>Digital therapeutics after Pear: separating the clinical evidence from the payment model</title><link>https://techanatomy.co/a/digital-therapeutics-pear-and-reimbursement.html</link><guid isPermaLink="true">https://techanatomy.co/a/digital-therapeutics-pear-and-reimbursement.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>Pear Therapeutics obtained three FDA authorisations for prescription software and filed for bankruptcy sixteen months after going public. The story separates two questions that are often merged: whether software can treat a disease, and who pays for it.</description></item><item><title>The European Health Data Space after HIMSS Europe 2026, and what Hal Wolf asks it to protect</title><link>https://techanatomy.co/a/european-health-data-space-hal-wolf.html</link><guid isPermaLink="true">https://techanatomy.co/a/european-health-data-space-hal-wolf.html</guid><pubDate>Sat, 07 Nov 2026 08:00:00 +0000</pubDate><category>Analysis</category><description>The regulation creating the European Health Data Space entered into force in March 2025, and its first obligations for patients and researchers apply from March 2029. Hal Wolf, president and CEO of HIMSS, explains why he sees its main task as keeping national health data silos from closing again.</description></item></channel></rss>
