Explorez tous les épisodes du podcast AI in Medicine - curated summaries making complex issues easy to understand

We unpack the FDA 510(k) summary for the Clarius Ejection Fraction AI, a machine learning tool that automates left ventricular ejection fraction measurements directly on handheld ultrasound probes. The hosts explore the clinical validation data, the performance metrics compared to expert human reviewers, and the implications of the software's Predetermined Change Control Plan for future AI updates. This episode highlights what the clearance means for clinicians at the bedside and the limitations inherent in the retrospective testing.

Key points

• Clarius Ejection Fraction AI is an assistive tool cleared for semi-automatic measurement of left ventricular ejection fraction on adult patients using specific Clarius ultrasound scanners.
• In a retrospective study of 279 exams, the AI demonstrated an Intraclass Correlation Coefficient of 0.78 when compared to the mean of human reviewers, whereas human reviewers scored between 0.54 and 0.67 when compared to each other.
• The clinical verification dataset was drawn from multiple geographic locations, but no clinical or sociodemographic information such as age, gender, or clinical diagnosis was available.
• Images of inadequate quality and subjects where the Ejection Fraction AI model failed to generate a measurement were excluded from the performance testing comparisons.
• The FDA clearance includes a Predetermined Change Control Plan (PCCP) allowing the company to modify data input sources, training hyperparameters, post-processing steps, and model architecture without a new 510(k) submission, provided they follow strict risk mitigation testing.

Source: FDA 510(k) summary K253593: Clarius Ejection Fraction AI (Clarius Mobile Health) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-clarius-ejection-fraction-ai-automating-cardiac-scan-772da0

The European Commission released its first formal assessment on whether to expand the list of prohibited and high-risk AI under the AI Act. We unpack the 547 stakeholder responses, the 3 791 real-world incidents analyzed, and what this means for digital health vendors navigating the gray area of mental health chatbots.

Key points

• The European Commission's first review concluded no immediate changes are needed to the lists of prohibited or high-risk AI, largely due to a lack of empirical enforcement data.
• Mental health and self-help chatbots were flagged by Member States as a complex area; depending on their features, they could fall under the Medical Devices Regulation, be classified as high-risk, or face outright prohibition.
• An analysis of 3 791 reported AI incidents revealed that approximately 10% are likely addressed by the AI Act's product safety rules in Annex I, while around 30% required detailed analysis for high-risk or prohibited status.
• The Commission identified a critical potential regulatory gap where AI generating non-consensual deepfakes and child sexual abuse material does not currently meet the strict legal definition of a prohibited manipulative practice under Article 5.
• For medtech and digital health companies, the message is clear: the risk tiers are set for now, but gray areas are being heavily monitored for future amendments in 2026.

Source: Report from the Commission to the European Parliament and the Council: Report adopted pursuant to Article 112(1) of Regulation (EU) 2024/1689 on the need to review the list of prohibited AI practices and of high-risk AI systems listed in Annex III - European Commission, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/drawing-the-red-lines-the-eu-s-first-formal-review-of-prohibited-and-h-811f72

Millions of patients receive intravenous therapy every day, yet the infusion pumps delivering these critical medications remain highly vulnerable to human error. This episode unpacks a comprehensive review on the convergence of wearable biosensors, smart infusion automation, and artificial intelligence to explore how hospitals can finally achieve autonomous, closed-loop fluid delivery. We discuss the current maturity of clinical-grade wearables, the prototype stage of smart pumps, and the massive interoperability and security gaps that medtech companies must solve to make proactive care a reality.

Key points

• A systematic review of 112 studies reveals a stark maturity gap: wearable biosensors have achieved near clinical-grade accuracy, while most smart IV systems remain in the prototype phase without closed-loop control.
• Current wearable technologies can measure vital signs with remarkable precision, demonstrating heart rate errors as low as ±1.1 bpm and peripheral oxygen saturation errors of ±0.72%.
• The proposed INFUSYNC framework aims to merge continuous patient monitoring, intelligent fluid delivery, and data-driven analytics into a single, synchronized healthcare system.
• Despite advancements in theoretical security frameworks like blockchain and federated learning, current IV infusion systems lack built-in data protection, highlighting a critical need for security-by-design.
• Computer vision models like the Pyramid Scene Parsing Network show superior accuracy over the Segment Anything Model in monitoring IV fluid levels under challenging conditions like surface reflections and label occlusion.
• The next frontier for medtech is moving beyond isolated, alert-based gadgets toward fully integrated systems that feature bidirectional communication with electronic health records via HL7 FHIR standards.

Source: Current status and future outlook of artificial intelligence and automation in intravenous infusion systems - NPJ digital medicine, 2026 (CC BY)

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/closing-the-loop-how-ai-and-wearables-are-rewriting-the-rules-of-iv-in-9c1c89

We unpack the FDA 510(k) summary for Cosmo AI's GI Genius Module 300 and ColonPRO US software. We explore its 88.07% lesion-based sensitivity and the newly cleared multicontainer functionality that allows non-medical health IT software to run in the background during live procedures.

Key points

• The GI Genius Module 300 is a computer-assisted reading tool designed to aid endoscopists in detecting colonic mucosal lesions during standard white-light endoscopy.
• The system demonstrated an 88.07% lesion-based sensitivity on a dataset of 150 videos containing 338 polyps.
• This Special 510(k) clearance introduces a multicontainer management functionality, allowing non-medical background software to run simultaneously with the medical lesion detection software.
• Hardware ports including LAN, USB, and QSFP, previously used only for manufacturer service, are now accessible for user interaction with these background applications.
• Verification testing confirmed that running background applications does not impact the timing, image processing, or 29 μs video delay of the foreground medical software.

Source: FDA 510(k) summary K261369: GI Genius™ Module 300 (GGM300-US);ColonPRO™ US (CPRO403S-US) (Cosmo Artificial Intelligence - Ai) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-cosmo-ai-gi-genius-module-300-multitasking-in-endosc-3ab1fd

The FDA is seeking input on a new pilot program exploring how artificial intelligence can optimize early-phase clinical trials. Maya and Sam unpack the agency's goals—from accelerating dose selection to guiding critical go/no-go decisions—and what these proposed standards mean for pharma sponsors and AI developers.

Key points

• The FDA identifies early-phase clinical trials as a critical bottleneck characterized by high uncertainty and inefficient decision-making processes.
• The proposed pilot program aims to explore how AI can improve trial efficiency, enhance safety monitoring, and support earlier Phase 1 to Phase 2 decisions.
• The FDA emphasizes that AI use must align with the NIST AI Risk Management Framework, ensuring systems are valid, safe, secure, explainable, and fair.
• The agency is actively soliciting public input on how to structure the pilot, including which therapeutic areas to target and how to measure AI system stability and model drift.
• Evaluating the concordance between AI-supported and traditional decision-making will be a key focus for sponsors and regulators alike.

Source: AI-Enabled Optimization of Early-Phase Clinical Trials Pilot Program; Request for Information - U.S. Food and Drug Administration

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/ai-at-the-bottleneck-the-fda-s-blueprint-for-smarter-early-phase-trial-85dd54

We unpack the FDA 510(k) summary for SubtleHD-CT, an AI software designed to enhance CT scans in the background. We explore the phantom testing parameters, the retrospective reader study results, and a crucial limitation regarding which image reconstruction types the software supports.

Key points

• SubtleHD-CT is an AI software cleared to process and enhance CT images regardless of the scanner manufacturer or model.
• The software runs entirely in the background without a user interface, automatically receiving images, enhancing them, and sending them to the hospital's PACS.
• Clearance was based on substantial equivalence to a predicate device, ClariCT.AI, and included extensive phantom testing across multiple scanner brands.
• A retrospective reader study of 270 clinical cases showed the enhanced images met the required thresholds for diagnostic utility and lack of artifacts.
• A major limitation noted in the document is that the software is engineered for standard Filtered Back Projection; use with Iterative Reconstruction or Deep Learning reconstructions may result in unknown performance.

Source: FDA 510(k) summary K254120: SubtleHD-CT (1.x) (Subtle Medical) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-subtle-medical-subtlehd-ct-upgrading-ct-image-qualit-cc13b9

We dive into the FDA 510(k) summary for iSchemaView's Rapid Aortic Measurements, an AI tool designed to evaluate aortic and iliac arteries. We explore its performance data, curious dataset discrepancies, technical limitations, and what this clearance means for health systems looking to automate vascular imaging workflows.

Key points

• Rapid Aortic Measurements (AM) is an AI software that segments and measures the aorta and iliac arteries on CT scans, acting as an aid to physician decision-making.
• The clearance is based on substantial equivalence to the previously cleared Rapid Neuro3D device, indicating a technological shift from brain imaging down to the chest, abdomen, and pelvis.
• Performance testing reported a 100% agreement on segmentation and clinical accuracy by up to 3 clinical experts across a dataset of 108 cases.
• The document contains unexplained dataset discrepancies, such as reporting 108 cases derived from 115 patients, and demographics that do not align perfectly.
• The software has strict exclusion criteria, including series with excessive motion, metal implants, slice thickness greater than 3 mm, and scans from cone-beam CT scanners.

Source: FDA 510(k) summary K251987: Rapid Aortic Measurements (iSchemaView) - U.S. Food and Drug Administration, 2025

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-ischemaview-rapid-aortic-measurements-the-shift-from-d7896a

This episode unpacks an executive case note from The AI Institute arguing that hospital boards must approve clinical AI and its operating environment as a single, joint decision. We explore real-world lessons from the UK's AI Airlock programme, the danger of splitting product and operational risk into separate rooms, and the four-part approval gate every health leader should implement before deployment.

Key points

• Treating product capability and operating readiness as one investment case is critical; a capable model with an untrained user is not the approved system.
• The UK's AI Airlock programme tested real AI products across seven challenge areas, revealing that local data, workflows, and user interfaces can change an algorithm's performance.
• Health leaders must name and fund the conditions the user environment must maintain before deployment, including cybersecurity, incident reporting, and service continuity.
• Material changes to a model, its data, the workflow, the user group, or the population require a formal reapproval trigger.
• While the principle of joint approval travels globally, local laws, professional accountability, and public trust requirements do not.
• A regulatory sandbox is a learning architecture, not a badge of success, and a recommendation is not a deployment outcome.

Source: Approve the AI and Its Operating Environment Together - The AI Institute, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/beyond-the-algorithm-why-buying-ai-without-the-operating-environment-i-956023

In this company spotlight, we break down the FDA 510(k) summary for Volta Medical's Volta AF-Xplorer II, cleared in September 2025. We explore how this AI tool uses deep learning to help electrophysiologists annotate 3D heart maps by analyzing intracardiac signals. Listeners will learn about its hardware requirements, catheter compatibility, algorithm updates, and the bench testing data that secured its substantial equivalence clearance.

Key points

• Volta AF-Xplorer II is an AI software and computing platform cleared by the FDA to assist in real-time annotation of 3D anatomical and electrical maps during atrial fibrillation or atrial tachycardia.
• The system analyzes 1.5 seconds of intra-cardiac electrogram recordings sampled at 1000Hz to detect spatiotemporal dispersion.
• Updates from the predicate device include an extended training dataset, new "state of the art classifiers", and a simplified visual output that includes a gauge for dispersion stability.
• The clearance is based on substantial equivalence to the original Volta AF-Xplorer, supported by unitary tests showing a 100% conformity rate and expert assessment of 2D maps in a simulated environment.
• The 510(k) summary does not include new prospective clinical trial data or specify the exact numerical size of the expanded training dataset.

Source: FDA 510(k) summary K251518: Volta AF-Xplorer II (Volta Medical) - U.S. Food and Drug Administration, 2025

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-volta-medical-volta-af-xplorer-ii-ai-for-atrial-mapp-eb23f2

The OECD's 2026 report outlines how healthcare can move AI from isolated pilot projects to enterprise-scale deployment. Listeners will learn the critical data, governance, and workforce strategies needed to overcome bottlenecks and safely scale AI across health systems.

Key points

• 100% of OECD member countries use AI in administration, but only 10% use medical imaging applications at national-level scale-up.
• While health data represents 30% of all data, the sector uses less than 5% of it for decision making.
• Only 11% of countries are updating national procurement guidelines, and 24% are streamlining health technology assessments to include AI.
• Ambient voice technology reduced documentation time by 85.8%, saving an average of 5.27 minutes per case.
• The OECD advocates for Health Data Access Bodies, living model cards, and cross-border regulatory compatibility to balance innovation with patient safety.

Source: Scaling Artificial Intelligence in Health - Organisation for Economic Co-operation and Development, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/escaping-pilot-purgatory-the-oecd-blueprint-for-scaling-ai-in-health-b35b6e

In this Company Spotlight, we explore the FDA 510(k) summary for Sonio Suspect, an AI tool designed to detect abnormal fetal ultrasound findings. We unpack how Sonio transitioned their algorithm from a cloud-based web application to an integrated, on-device solution, examining the performance data, the exact anatomical anomalies it screens for, and what this regulatory shift means for medtech developers and clinicians.

Key points

• Sonio Suspect is an AI software cleared to assist interpreting physicians in identifying abnormal fetal ultrasound findings from 11 weeks to 41 weeks of gestational age.
• This specific 510(k) clearance allows Sonio to shift from a Python-based cloud application to a C++ algorithm integrated directly into third-party ultrasound systems.
• The integrated device demonstrated an average sensitivity of 92.5% and specificity of 90.4% across tested findings, establishing substantial equivalence to the predicate device.
• The testing validated performance across specific chest, abdominal, and cephalic anomalies, though the public summary does not disclose the total number of scans in the test dataset.
• The FDA document emphasizes that patient management decisions should not be made solely on the results of the Sonio Suspect analysis.

Source: FDA 510(k) summary K261519: Sonio Suspect (Sonio) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-sonio-suspect-moving-fetal-ultrasound-ai-from-the-cl-3260c2

NICE recently published guidance on 10 artificial intelligence technologies designed to detect or characterise colorectal polyps. While 6 tools have been conditionally recommended for NHS use, the economic and clinical real-world impact remains uncertain. We unpack what this means for gastroenterologists, the 4-year evidence generation plan, and why finding more polyps does not automatically equal preventing more cancer.

Key points

• NICE recommends 6 AI technologies for use in the NHS during a 4 years evidence generation period for people without diagnosed IBD or Lynch syndrome.
• Clinical evidence shows AI improves the adenoma detection rate overall, but mostly for smaller polyps (less than 5 mm) rather than advanced adenomas.
• The cost-effectiveness of these tools is highly uncertain, hovering at a 50% probability at NICE's willingness-to-pay range of £20,000 to £30,000.
• More research is required for AI tools that characterise polyps and for using AI in patients with inflammatory bowel disease or Lynch syndrome.
• Increased detection of benign polyps could lead to uncaptured downstream costs, including more histology testing and unnecessary surveillance colonoscopies.

Source: Artificial intelligence (AI) technologies to help detect or characterise colorectal polyps - National Institute for Health and Care Excellence

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/will-ai-find-the-right-polyps-decoding-nice-guidance-on-colonoscopy-to-619f4d

How do you evaluate a medical technology that constantly changes after deployment? Maya and Sam unpack a new report from the University of Sheffield's Decision Support Unit on how health technology assessment bodies like NICE must adapt to 'frontier AI'. They explore the nightmare of evidence decay, the hidden costs of governance, and the new requirements medtech companies will face to get AI funded.

Key points

• Frontier AI is defined by its nature as a configurable, evolving service rather than a fixed product, causing constant intervention instability.
• Health technology assessments face rapid evidence decay because routine updates to AI models and prompt templates can alter safety and performance.
• Economic models must evolve beyond software licensing fees to include the massive ongoing costs of cybersecurity, workflow integration, and continuous governance.
• NICE may require Predetermined Change Control Plans to ensure medtech companies outline exactly how post-deployment updates will be managed and tested.
• Recommendations for frontier AI will likely be bound to specific versions, configurations, and strict scopes of use, with mandatory review triggers if the system changes.

Source: FRONTIER ARTIFICIAL INTELLIGENCE AND HEALTH TECHNOLOGY ASSESSMENT - Decision Support Unit, University of Sheffield, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/when-ai-changes-overnight-how-nice-is-rewriting-the-rules-for-evaluati-bef9c8

We unpack the FDA 510(k) clearance for Anumana's WatchMate Software, a deep-learning tool that reconstructs 3D models of the left atrial appendage from standard 2D ultrasound. Discover how it stacks up against human physicians and CT scans, its clinical trial results, and the practical IT realities of deploying AI in the cath lab.

Key points

• WatchMate Software is cleared for pre-procedural planning and device sizing for left atrial appendage closure (LAAC) procedures.
• It uses a multi-stage AI pipeline to reconstruct a 3D model and estimate a landing zone from 2D transesophageal echocardiography (TEE) images.
• In a 100-subject prospective study, WatchMate achieved a mean absolute deviation of 3.1 ± 3.4 mm compared to a CT reference standard, outperforming the human 2D TEE average of 5.2 ± 3.4 mm.
• The software operates locally on a Linux workstation with an NVIDIA GPU and an iPad mobile interface, completely bypassing cloud connectivity.
• Clearance is based on substantial equivalence to the TruPlan CT Imaging Software, highlighting a shift from CT-based planning to TEE-based AI planning.
• Real-world limitations apply: 100 subjects were enrolled, but only 72 were included in the per-protocol population due to unanalyzable data or 2D-3D inconsistency flags.

Source: FDA 510(k) summary K260300: WatchMate Software (Anumana) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-anumana-watchmate-turning-2d-echo-into-3d-surgical-p-9a3d58

We unpack the Stanford AI Index 2026 Medicine chapter to explore why AI is acing medical exams but still struggling to translate into prospective clinical trial evidence. We dive into the massive gap between FDA clearances and randomized trials, highlight which tools like ambient scribes are genuinely scaling, and reveal why smaller models are outperforming massive systems in molecular biology.

Key points

• Leading multi-agent AI systems score 85.5% on complex medical cases compared to 20% for physicians, yet nearly half of clinical AI studies still rely on simulated scenarios rather than real patient data.
• Ambient AI scribes are driving real clinical impact, with some health systems reporting an 83% reduction in note-writing effort and up to a 112% return on investment.
• The FDA authorized 258 AI medical devices in 2025, but a peer-reviewed analysis found that only 2.4% of devices with clinical studies were supported by randomized controlled trial data.
• In molecular biology, smaller specialized models are outperforming massive general-purpose systems, and biological model development is increasingly bottlenecked by data curation rather than architecture.
• AI-generated summaries now appear at the top of 84% to 92% of health-related Google searches, fundamentally shaping how patients access health information before ever seeing a clinician.

Source: Medicine - Stanford Institute for Human-Centered Artificial Intelligence, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/exam-scores-vs-reality-the-2026-ai-index-report-on-medicine-ec9148

We unpack the FDA 510(k) summary for Bunkerhill Contrast CAC, an AI tool cleared to opportunistically detect and quantify coronary artery calcium on routine contrast-enhanced chest CTs. We discuss the shift from non-contrast to contrast imaging, the pivotal performance data, and what it means for clinical workflows.

Key points

• Cleared for adults aged 30 years and above to detect and estimate the quantity of coronary artery calcification.
• Specifically designed to analyze non-gated, contrast-enhanced chest CT images, unlike its predicate device which uses non-contrast scans.
• Outputs an estimated volume score in mm3 rather than traditional Agatston units.
• Pivotal testing demonstrated a sensitivity of 0.957 and specificity of 0.929.
• A 50-case bridging study comparing contrast to non-contrast images showed a mean bias of -4.36 mm3.
• The software is intended as adjunctive information only and does not replace the physician's evaluation or the original patient report.

Source: FDA 510(k) summary K260166: Bunkerhill Contrast CAC (BunkerHill Health) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-bunkerhill-contrast-cac-opportunistic-screening-on-r-dfb2f3

The FDA is asking the healthcare industry how to regulate generative AI medical devices. We unpack their proposed competency-based approach, which evaluates AI less like software and more like a human medical resident.

Key points

• Generative AI requires new regulatory frameworks because it produces variable outputs to similar inputs and can evolve after deployment.
• The FDA proposes a two-axis framework to assess risk based on device activity (how independently it acts) and the consequences of an incorrect output.
• A new competency-based premarket evaluation would mirror human clinician credentialing, using robust benchmarking and clinical confirmation.
• Postmarket monitoring could involve periodic re-benchmarking and shared responsibility across the healthcare ecosystem to catch performance degradation.
• The FDA suggests voluntary master files for foundation models, allowing developers to share underlying safety data without making it public.

Source: Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/credentialing-the-ai-doctor-the-fda-s-blueprint-for-generative-ai-944c85

We unpack the FDA 510(k) summary for Tyto Insights for Wheeze Detection (with PCCP). Listeners will learn how this over-the-counter AI software analyzes lung sounds, the real performance numbers behind it, and exactly how the FDA's Predetermined Change Control Plan allows algorithms to be retrained post-clearance without new submissions.

Key points

• The software is an over-the-counter AI tool that analyzes 6 to 12-second acoustic lung signals from a compatible stethoscope to detect suspected wheezing in patients 2 years and older.
• Performance testing showed a specificity of 0.99 and a sensitivity of 0.55, meaning it rarely flags false positives but misses a significant portion of actual wheezes.
• This 510(k) clearance is based on substantial equivalence to a 2023 version of the same product, with the changes being cybersecurity updates and the addition of a Predetermined Change Control Plan (PCCP).
• The PCCP allows the company to retrain the machine learning model to improve performance without filing a new 510(k), provided the new version's sensitivity and specificity do not drop below the currently validated thresholds.
• The FDA approval letter emphasizes that even with a PCCP, all software modifications must still follow strict Quality Management System Regulations, including design controls and documentation in the Medical Device File.

Source: FDA 510(k) summary K252844: Tyto Insights for Wheeze Detection (with PCCP) (Tyto Care) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-tyto-care-tyto-insights-the-fda-s-blueprint-for-upda-ac2251

The AMA just surveyed nearly 1,700 physicians on their real-world use of augmented intelligence. We unpack why professional adoption has doubled, the massive hidden wave of patient AI use, and the specific liability fears keeping doctors awake at night. Essential listening for clinical leaders and medtech product teams.

Key points

• Over 80% of physicians now use AI professionally, averaging 2.3 use cases, with research summarization leading the pack at nearly 40%.
• While 76% see AI as an advantage for patient care, 88% are concerned about clinical skill loss, particularly among medical students and residents.
• There is a massive disclosure gap: 30% of doctors assume most of their patients use AI for health questions, but 29% haven't had a single patient disclose it.
• "Clear liability frameworks" ranks as the number one regulatory priority, with 88% demanding validation of safety and efficacy before adoption.
• Despite 92% of physicians wanting more AI training, surprisingly, physicians under 35 report lower training rates (55%) than those over 55 (71%).

Source: 2026 Physician Survey on Augmented Intelligence - American Medical Association, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/the-ai-exam-room-what-1-692-physicians-actually-want-and-fear-in-2026-35217f

We unpack the FDA 510(k) summary for Eko Health's EFAST software, which brings transformer-based AI to the digital stethoscope. Listeners will learn how the cloud-based system analyzes heart sounds and ECGs, its real-world performance numbers for detecting structural murmurs and atrial fibrillation, and the crucial caveats regarding pediatric screening and retrospective testing.

Key points

• EFAST is a cloud-based Software as a Medical Device (SaMD) that analyzes 15-second heart sound and ECG recordings via an API, returning a JSON format data structure rather than using a primary display.
• For structural murmur classification, EFAST showed an overall sensitivity of 83.4% and specificity of 86.0%, closely matching its predicate device.
• Specificity for pediatric patients dropped to 76.0%, which the document notes is likely due to confusion with innocent murmurs.
• For atrial fibrillation detection in adults, sensitivity leaped to 94.7% (compared to 73.1% in the older reference device), with a specificity of 94.1%.
• Testing was retrospective, and experts excluded 259 out of 2,460 murmur recordings and 97 out of 1,256 ECG recordings because they could not come to an agreement.
• Both test datasets heavily skewed White (90.0% for murmurs, 86.6% for ECG), which the FDA summary describes as representative of the intended use population.

Source: FDA 510(k) summary K251494: Eko Foundation Analysis Software with Transformers (EFAST) (Eko Health) - U.S. Food and Drug Administration, 2025

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-eko-health-efast-transformers-hit-the-digital-stetho-989647

We dive into the FDA's recent clearance of AiTiALVSD, an AI tool that predicts left ventricular systolic dysfunction from a standard 12-lead ECG. We break down the clinical validation numbers, the demographic gaps in the training data, and what this specific 510(k) clearance means for cardiology workflows.

Key points

• AiTiALVSD by Medical AI Co., Ltd. received FDA Class II clearance to support the diagnosis of Left Ventricular Systolic Dysfunction (LVEF less than or equal to 40%) using 12-lead ECG data.
• The deep learning model was trained on 400,339 ECGs from 148,624 patients across 16 hospitals in Korea, and evaluated in a US retrospective study of 22,101 patients.
• In the US performance study, the software demonstrated a sensitivity of 83.7 and a negative predictive value of 97.7, making it a strong potential rule-out tool for clinicians.
• The FDA document explicitly notes a limitation regarding Black or African American subjects, who made up 9.6% of the validation study, noting that specificity was lower in this subgroup.
• The software provides a binary output of High or Low, but the FDA strictly warns it must not be used as a substitute for echocardiography or as a standalone diagnosis.

Source: Re: K254302 - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/fda-clears-ai-for-heart-failure-unpacking-the-aitialvsd-510-k-8822ad

We unpack a critical Stanford and UC Berkeley report advising California on its new Transparency in Frontier AI Act (TFAIA). Discover why current legal definitions of AI risk are dangerously narrow, how a strict focus on computing power ignores biological threats, and what new reporting thresholds could mean for healthcare and medtech developers.

Key points

• A computing threshold of 10^26 operations is an imperfect proxy for risk, as smaller models trained on biological data can still pose catastrophic threats.
• The original law defines catastrophic risk as 50 deaths or 1,000,000,000 dollars in property damage, which fails to capture cumulative harms like mental health deterioration.
• Basing compliance on 500,000,000 dollars in revenue misses highly funded startups; the report suggests adding a 100,000,000 dollars R&D spending threshold.
• Internal testing incidents, such as loss of control within a developer's sandbox, should be reportable even if they do not cause immediate physical harm.
• Meaningful AI evaluation requires a sociotechnical approach that looks beyond the model to the tools, workflows, and human interactions surrounding it.

Source: Implementing the Transparency in Frontier AI Act (TFAIA) Report for the California Department of Technology Multistakeholder Workshops - Stanford Institute for Human-Centered AI

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/when-ai-escapes-the-lab-redefining-catastrophe-in-california-s-frontie-d9e2fd

NICE weighs in on 4 AI technologies designed to speed up echocardiography for heart failure diagnosis. Despite incredible claims of time savings, the agency stops short of recommending NHS funding. Maya and Sam unpack the evidence gaps, the economic models, and what AI builders need to prove before winning health-system adoption.

Key points

• NICE published an early-use assessment in May 2026 on 4 AI echocardiography tools: EchoConfidence, EchoGo Heart Failure, Ligence Heart, and Us2.ai.
• None of the technologies are recommended for NHS funding yet; access should remain through company, research, or non-core funding.
• While AI reduced analysis time in some studies—from 587 seconds to 3.2 seconds—it is uncertain if this translates to reduced waiting lists in the real world.
• The clinical evidence lacked generalisability, with many studies being retrospective, outside the UK, and excluding poor-quality scan images.
• Economic modeling showed potential cost savings or acceptable cost-effectiveness, but results are highly uncertain without real-world procedure time data.

Source: Artificial intelligence (AI)-assisted echocardiography analysis and reporting to support the diagnosis and monitoring of heart failure: early-use assessment - National Institute for Health and Care Excellence

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/can-ai-clear-the-echocardiogram-backlog-nice-s-verdict-on-heart-failur-95f0af

We unpack the FDA 510(k) summary for icobrain aria by icometrix, an AI tool cleared to help radiologists detect brain swelling and bleeding in Alzheimer's patients. We explore the clinical trial data, the significant jumps in sensitivity for spotting mild abnormalities, and what it means for the rollout of new amyloid beta-directed therapies.

Key points

• icobrain aria is a software-only device that assists radiologists in detecting ARIA (amyloid-related imaging abnormalities) from MRI scans of Alzheimer's patients.
• In a clinical study with 16 radiologists, the software increased sensitivity for detecting mild ARIA-E from 47.2% to 70.2%.
• The software provides color-coded 3D overlays and a two-page electronic report, tracking changes in edema and microhemorrhages across two time points.
• It does not replace clinical judgment and cannot segment macrohemorrhages of 10 mm or more.
• FDA clearance was based on substantial equivalence to OsteoDetect, highlighting how AI in radiology is evaluated on workflow and algorithmic similarities rather than identical disease states.

Source: FDA 510(k) summary K240712: icobrain aria (icometrix NV) - U.S. Food and Drug Administration, 2024

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-icometrix-icobrain-aria-the-ai-unlocking-alzheimer-s-fcc19a

MedTech Europe's position paper outlines a critical juncture for medical device regulation in the EU. We unpack their proposed amendments to the MDR and IVDR, focusing on how AI should be regulated, the fight over cybersecurity reporting, and why one in three manufacturers is deprioritising the EU for product launches.

Key points

• One in three manufacturers is currently deprioritising the European Union for first regulatory approvals due to administrative burden and costs.
• MedTech Europe strongly supports embedding AI Act requirements into the MDR and IVDR to allow for a single conformity assessment, preventing dual-regulatory overlap.
• They propose a 12-month deadline for the European Commission to adopt delegated acts integrating high-risk AI requirements into existing safety frameworks.
• The position paper calls for structurally separating cybersecurity vulnerability reporting from traditional patient safety vigilance to avoid false alarms and regulatory confusion.
• For in vitro diagnostics, MedTech Europe wants the orphan device threshold changed from 1 in 12,000 to 5 in 10,000 to align with the established rare disease definition.
• The industry opposes a proposed regulatory shift that would make devices reusable by default, arguing that single-use engineering is fundamental to patient safety.

Source: MDR/IVDR Revision: A regulatory system at a crossroads - MedTech Europe, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/will-the-ai-act-break-european-medtech-inside-the-mdr-ivdr-revision-5f21c6

The CDC just released a blueprint for how field epidemiology training programs should handle AI. We explore a fascinating case study where AI actually harms an outbreak investigation, and unpack the new rules for training the next generation of disease detectives to use predictive models safely.

Key points

• AI models predict sequences based on probability, which can mimic human thinking but bypass the critical cognitive habits trainees need to develop expert judgment.
• Over-reliance on AI can give new learners a false appearance of competence by generating polished deliverables before foundational skills are internalized.
• A detailed case study demonstrates how an AI's confident, authoritative list of plausible pathogens can anchor an investigator and lead to premature closure in a novel outbreak.
• The CDC recommends a human-in-the-loop approach, where trainees verify all facts, check statistical assumptions, and never use AI to entirely draft manuscripts or unguided data analyses.
• Training programs must adapt by incorporating oral assessments and proctored simulations to ensure epidemiologists can independently reason through messy, real-world data.

Source: Implementing artificial intelligence (AI) in field epidemiology training programs (FETPs) - Centers for Disease Control and Prevention, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/who-solves-the-outbreak-when-ai-is-wrong-the-cdc-s-rules-for-disease-d-559667

The World Bank takes a deep dive into how artificial intelligence is actually being used in medical schools across Viet Nam. Maya and Sam unpack the massive gap between student enthusiasm and institutional readiness, the risks of informal AI adoption, and what it means for the future of clinical training globally.

Key points

• 81.1 percent of medical students surveyed have used an AI-powered tool, but mostly for individual productivity, not through institutional platforms.
• Nearly half of medical students and teachers have never received formal training on how to use AI.
• A major barrier to advanced AI integration is 'digital data underdevelopment', with many institutions still relying on paper-based records.
• While over 90% of students believe AI will be essential in the future, significant concerns remain around academic integrity, over-reliance, and reduced critical thinking.
• There are stark demographic divides in AI usage, with younger faculty and male students adopting the technology at significantly higher daily rates.
• Without formal curricula, students rely on social media and free internet tools to learn AI, creating equity and quality concerns.

Source: ARTIFICIAL INTELLIGENCE INTEGRATION IN MEDICAL EDUCATION: A Landscape with HOT Dimensions in Viet Nam - The World Bank, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/will-ai-replace-the-medical-curriculum-inside-the-world-bank-s-latest-e85848

We unpack the FDA clearance letter for ArteraAI Breast, a software-only device designed to predict 5- and 10-year metastasis risks from standard pathology slides. We explore the specific patient population it targets, the regulatory significance of its Predetermined Change Control Plan, and why an administrative clearance letter doesn't include performance data.

Key points

• ArteraAI Breast is an FDA-cleared, software-only device that analyzes whole slide images of H&E stained tissue alongside patient age, tumor size, and nodal status.
• It outputs 5- and 10-year risks of distant metastasis and an ArteraAI risk score for a specific cohort: adult patients with early-stage, HR+/HER2-, N0 or N1 invasive breast cancer.
• The FDA clearance includes a Predetermined Change Control Plan, allowing the company to make certain device modifications without a new 510(k) submission.
• Because this document is the administrative clearance letter and not the full 510(k) summary, it does not name the predicate device or provide performance testing numbers.
• The clearance mandates strict adherence to the Quality Management System Regulation, including specific ISO 13485 clauses for design controls and corrective actions.

Source: FDA 510(k) summary K254115: ArteraAI Breast (Artera) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-arteraai-breast-predicting-metastasis-from-h-e-slide-31d8ec

The World Economic Forum has released a new framework on how Asia is deploying AI. We explore why rapid AI adoption is failing to translate into sustained value for many organizations, and how health systems and medtech companies must redesign human roles around "direction, judgement and accountability" to fix this bottleneck.

Key points

• While 77% of organizations adopt advanced AI, very few report achieving widespread and sustained value.
• Asia is a crucial testing ground for AI scale, accounting for 74% of new industrial robot deployments in 2024 and nearly 59% of the global population.
• The bottleneck in AI transformation is no longer model capability, but the widening gap between technical advancement and the readiness of human systems to adapt.
• Organizations must design human roles around 3 non-delegable functions: direction setters, judgement keepers, and accountability holders.
• Different Asian economies are taking distinct pathways: China as a broad-based scaler, Japan as a sequenced adopter, and India as a sector-priority adopter.
• Only 8% of surveyed APAC organizations have adjusted job roles or decision responsibilities for AI, highlighting a dangerous trend of deployment before design.

Source: Asia’s Human-led AI Opportunity: A Framework for Transformation - World Economic Forum, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/human-led-ai-in-asia-why-adoption-is-outpacing-value-3d9ea7

MedTech and Pharma companies are using AI to speed up regulatory submissions, but many are finding that faster output is actually leading to more rejections. This episode unpacks a 2026 white paper from IQVIA detailing why speed-first AI fails by treating regulations like a checklist, missing crucial cross-functional inconsistencies. Listeners will learn why shifting from pure speed to patient-centric velocity—and embedding AI directly into regulatory workflows—is the only way to build defensible submissions that regulators can trust.

Key points

• AI optimized solely for throughput validates completeness rather than correctness, accelerating mistakes and hallucinations.
• Speed-first AI systems miss cross-functional inconsistencies between clinical evidence, technical data, quality, and labeling claims.
• Global regulators now have access to their own AI tools, allowing them to spot disjointed, speed-driven submissions instantly.
• IQVIA argues for a shift to velocity—speed combined with intentional, patient-centric direction grounded in clinical evidence.
• AI should be embedded directly into Regulatory Information Management workflows rather than layered on top, catching potential objections before submission.
• Accountability remains with human regulatory professionals, who must ensure the submission tells a clear, defensible story about patient safety.

Source: Speed Can Mislead: Why Velocity Matters Most in Global Regulatory Affairs - IQVIA, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/why-faster-ai-submissions-mean-slower-approvals-iqvia-on-regulatory-ve-0bafdc

The FDA recently cleared TumorSight Risk, an AI software that analyzes MRI and clinical data to predict the 5- and 10-year recurrence risk for breast cancer patients. Maya and Sam unpack the 510(k) summary, exploring the algorithm's impressive negative predictive value, its retrospective validation on 2,129 patients, and what it means for the future of prognostic AI in clinical practice.

Key points

• TumorSight Risk predicts 5- and 10-year recurrence risks for patients with early-stage, HR+/HER2- breast cancer using a proprietary score from 0-100.
• The model demonstrated a 97.0% negative predictive value at 5 years, meaning patients classified as low risk have a high probability of remaining recurrence-free in that window.
• Validation relied on a retrospective study of 2,129 patients diagnosed between 2010 and 2024, proving the need for long-term historical data in prognostic AI.
• The software uses Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) alongside clinical inputs like age, race, and tumor grade.
• The FDA cleared it as a Class II device by finding it substantially equivalent to ArteraAI Breast, even though the predicate used pathology slides rather than MRI data.

Source: TumorSight Risk - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/predicting-breast-cancer-recurrence-inside-the-fda-clearance-of-tumors-ae3f9f

DeepHealth’s Saige-Dx recently received FDA 510(k) clearance for analyzing digital breast tomosynthesis (DBT) mammograms. We unpack the public summary to see how this AI tool uses prior screening exams to reduce false positive alerts, and break down the exact testing demographics behind the clearance.

Key points

• Saige-Dx is an AI software cleared as a concurrent reading aid to identify soft tissue lesions and calcifications in DBT mammograms.
• This new clearance expands upon the predicate device by adding compatibility with Siemens systems and the ability to process prior exams.
• Processing prior exams was validated to reduce false positive marks per image by dropping false positive bounding boxes when a matching prior finding is identified.
• Performance was evaluated in two retrospective, standalone studies involving 2,425 cases and 684 cases, proving non-inferiority in AUC.
• The FDA clearance strictly labels Saige-Dx as an adjunct tool, warning that decisions should not be made solely based on its analysis.

Source: FDA 510(k) summary K253825: Saige-Dx (DeepHealth) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-deephealth-saige-dx-using-prior-exams-to-drop-false-6031cd

The FDA recently cleared the Artificial Intelligence Renal Assessment AIRA Version 1 for measuring kidney volume from CT scans. We break down the device's technical workflow, its impressive clinical performance metrics, and the strict regulatory controls required for AI software in medical imaging.

Key points

• AIRA Version 1 received Class II clearance for automatic image segmentation and calculation of renal parenchymal volume using CT scans.
• The algorithm's performance was validated in a retrospective standalone study of 113 cases, achieving a mean Dice Similarity Coefficient of 0.9092 for the right kidney and 0.9114 for the left.
• The study intentionally enriched its sample so that at least 25% of cases featured atrophied kidneys, ensuring the tool works on diseased anatomy.
• AIRA demonstrated substantial equivalence to the predicate device, Ceevra Reveal 3, but opted for a narrower focus: CT imaging only and limited to the kidney.
• The FDA clearance underscores the heavy quality management and regulatory requirements for Software as a Medical Device, including strict ISO 13485 design controls.

Source: Artificial Intelligence Renal Assessment (AIRA) Version 1 - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/unpacking-the-fda-clearance-of-the-aira-kidney-algorithm-ed2d94

This episode explores a pivotal paper on the shift from narrow AI tools to feedback-driven AI agents in oncology. We discuss how these systems orchestrate multimodal data, the current evidence from real-world and simulated studies, and what hospital leaders need to know about implementation risks and governance.

Key points

• AI agents differ from foundation models by using a feedback-driven loop to maintain state, select tools, and revise plans under constraints.
• Current evidence heavily supports component-level infrastructure, such as data layers and diagnostic modules, but prospective clinical validation of full agents is still lacking.
• Real-world data integration remains complex; one study successfully linked data for over 170,000 patients across 11 cancer types, yet local mapping and governance are always required.
• Clinical translation demands strict operational resilience, including explicit latency budgets, bounded retries, and clinician authority over final decisions.
• The most viable near-term application is transparent and traceable clinical decision support, particularly for multidisciplinary team case preparation, rather than autonomous diagnosis.

Source: AI Agents for Multimodal Oncology Diagnosis: Toward Transparent and Traceable Clinical Decision Support - JMIR cancer, 2026 (CC BY)

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/beyond-algorithms-how-ai-agents-could-reshape-multimodal-cancer-diagno-559a93

Obstructive sleep apnea affects an enormous global population, but traditional sleep studies are expensive and inaccessible. Maya and Sam unpack a new meta-analysis from the Journal of Medical Internet Research evaluating whether AI-powered photoplethysmography—the optical tech in many wearables—can accurately diagnose the condition. They explore the real-world accuracy of these models, the difference between deep learning and traditional machine learning, and what this means for the future of digital health screening.

Key points

• AI models trained on photoplethysmography (PPG) achieved a pooled sensitivity of 79.6% and specificity of 76.5% for diagnosing obstructive sleep apnea.
• Deep learning models significantly outperformed traditional machine learning in specificity, achieving 82.9% compared to 63.6%.
• As sleep apnea severity increases, the AI's specificity improves (up to 85.1%), but its sensitivity decreases (down to 76.7%), complicating its use across the disease spectrum.
• The type of device used—whether a wearable smart device or a clinical oximeter—was not clearly associated with a difference in diagnostic accuracy.
• While promising as a low-cost screening tool, the technology requires further validation with diverse, real-world data before it can replace conventional polysomnography.

Source: Accuracy of Artificial Intelligence in Diagnosing Obstructive Sleep Apnea Using Photoplethysmography: Systematic Review and Meta-Analysis - Journal of medical Internet research, 2026 (CC BY)

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/can-your-smartwatch-spot-sleep-apnea-unpacking-ai-and-photoplethysmogr-5b68b6

We dive into a comprehensive systematic review examining the real-world effectiveness, safety, and workflow burden of LLM-based medical report generation. Listeners will learn why highly rated AI drafts often increase editing times, why benchmark metrics obscure clinically significant errors, and why autonomous AI is not ready to replace clinician documentation.

Key points

• An AI impression-drafting study showed that editing AI drafts took longer (18.29 s) than the radiologist baseline (12.20 s) and increased edit distance.
• Out of 101 included studies evaluating AI report generation, 0 were judged at a low risk of bias.
• While an AI system achieved 70.5% acceptance in a chest x-ray study, it produced slightly higher false-negative rates (18.5%) than human radiologists (17.8%).
• Benchmark technical metrics like BLEU and ROUGE fail to capture clinically significant omission and commission errors.
• The evidence base is currently too heterogeneous and biased to support a pooled meta-analysis or any autonomous clinical-readiness claims.

Source: Effectiveness, Safety, and Workflow Burden of Large Language Model-Based Medical Report Generation: Systematic Review - Journal of medical Internet research, 2026 (CC BY)

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/do-ai-report-generators-actually-save-doctors-time-the-18-29-second-re-ac4bae

We unpack the FDA 510(k) summary for Galen Second Read by Ibex Medical Analytics. This episode explores how this cloud-hosted AI acts as a background safety net for pathologists, analyzing prostate biopsies initially diagnosed as benign to catch missed cancers, and what the real-world trade-offs in specificity look like.

Key points

• Galen Second Read is cleared as a Class II software algorithm device to assist users in digital pathology.
• It operates in the background specifically on prostate core needle biopsies initially diagnosed as benign by a pathologist.
• In a retrospective clinical study of cases initially diagnosed as benign, pathologists using the software achieved a sensitivity of 36.3%, catching cancers that would have otherwise been missed.
• The safety-net approach comes with a trade-off: overall specificity decreased by 3.2% when pathologists used the device compared to standard of care.
• The software is cloud-hosted, requires a minimum 20 Mbps internet connection, and is currently cleared for use with the Philips Ultra Fast Scanner.

Source: FDA 510(k) summary K241232: Galen™ Second Read™ (Ibex Medical Analytics) - U.S. Food and Drug Administration, 2025

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-ibex-galen-second-read-catching-missed-prostate-canc-9b7841

This episode unpacks the FDA 510(k) summary for Brainomix 360 Hyperdensity, an AI tool designed to automatically flag and measure intracranial hyperdensities on CT scans. We explore its clinical workflow, its performance metrics, and a fascinating pre-approved plan that allows the algorithm to be updated without new FDA submissions.

Key points

• Brainomix 360 Hyperdensity is a software-only medical device cleared to automatically label, visualize, and quantify intracranial hyperdensities on non-contrast CT head scans.
• The target population is adults who undergo a head scan for suspected acute intracranial hemorrhage as part of a primary diagnostic workup.
• The FDA cleared it based on substantial equivalence to the Viz HDS device, though Brainomix chose not to include outputs for lateral ventricle volume or midline shift.
• Performance testing showed a mean absolute error of less than 7.5 mL for total volume and a DICE score greater than 70%, though the summary omits the exact size of the test dataset.
• The clearance includes a Predetermined Change Control Plan, allowing Brainomix to retrain model weights and adjust probability thresholds in the future without a new premarket notification.

Source: FDA 510(k) summary K260406: Brainomix 360 Hyperdensity (Brainomix) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-brainomix-360-hyperdensity-automating-brain-bleed-me-d66dc3

A new World Economic Forum white paper argues that health systems are evaluating AI backward. Procurement teams buy AI for efficiency, while clinicians and patients bear the risks of unproven outcomes. Maya and Sam unpack why redefining value is the only way to win the health AI race.

Key points

• Global AI healthcare investment is breaking records, yet worker confidence fell by 18%.
• A structural value inversion means AI is currently evaluated on proximal metrics like convenience and revenue rather than meaningful clinical outcomes.
• Real-world evidence shows AI can cause clinician deskilling, and automation bias persists even after a 20-hour literacy program.
• Deploying AI requires context-sensitive frameworks, accounting for environments of scarcity where the alternative to AI is no care at all.
• The World Economic Forum proposes four principles to align developers, providers, and payers around patient-centric outcomes.

Source: Meaningful Outcomes Determine the Winners of the Health AI Race - World Economic Forum, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/the-value-inversion-why-meaningful-outcomes-will-decide-the-health-ai-ffd20d

Health systems are eager to adopt AI for aging populations, but does the technology actually save money? We unpack a scoping review of 40 economic evaluations to reveal why most AI tools are deemed cost-saving on paper, but often lack the real-world validation to prove it.

Key points

• Of 40 economic evaluations, 21 reported AI interventions to be cost-saving and 17 reported them to be cost-effective.
• Despite high methodological scores, model validation was not reported in 21 of the 40 included studies.
• Evidence is heavily skewed toward screening and diagnosis (32 out of 40 studies), with very few evaluating long-term monitoring or personalized care.
• AI-related costs were frequently derived from assumptions, expert opinion, or manufacturer quotes rather than real-world trial data.
• Key drivers of cost-effectiveness included algorithm performance, AI-related costs, underlying disease epidemiology, and local contextual factors like labor costs.

Source: Assessing the Value for Money of AI-Assisted Technologies for Older Adults: Scoping Review of Economic Evaluations - Journal of medical Internet research, 2026 (CC BY)

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/does-ai-in-eldercare-actually-save-money-a-hard-look-at-the-health-eco-84c385

This episode unpacks a comprehensive review spanning ten years of AI in screening mammography. Listeners will learn why pooled accuracy figures mask deep implementation challenges, how AI is effectively reducing reading workloads by up to 44.2%, and why local calibration is far more critical than an algorithm's out-of-the-box claims.

Key points

• Standalone AI accuracy pooled an AUC of 0.890 across over 1.2 million examinations, but the prediction interval was so wide (0.731–0.960) that out-of-the-box performance cannot be assumed.
• A bivariate model showed a summary sensitivity of 73.3% and a specificity of 92.4%, highlighting that algorithms miss roughly one in four screen-relevant cancers and should not be used as autonomous replacements.
• The landmark MASAI randomized trial demonstrated a 44.2% reduction in screen-reading workload alongside a cancer detection rate ratio of 1.29.
• Non-randomized implementations showed higher detection gains (pooled ratio 1.22), but these are highly vulnerable to confounding factors like self-selection bias.
• System updates and hardware changes severely impact AI performance; one UK program saw recall rates jump threefold simply due to a mammography equipment software upgrade.

Source: Ten Years of Artificial Intelligence in Screening Mammography: A Systematic Review and Meta-Analysis of Diagnostic Accuracy and Clinical Implementation (Literature Published 2015-2025) - Diagnostics (Basel, Switzerland), 2026 (CC BY)

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/mammography-s-ai-decade-from-retrospective-illusions-to-randomized-rea-1566ad

The UK government has officially accepted all 44 recommendations from the National Commission into the Regulation of AI in Healthcare. In this episode, Maya and Sam unpack the MHRA's extensive response, detailing the shift from point-in-time software checks to continuous, lifecycle-based regulation. They explore what this means for medical device manufacturers building adaptive algorithms, how healthcare systems will manage the shifting lines of clinical liability, and the practical steps the government is taking to deploy a massive tech investment safely.

Key points

• The UK government accepted all 44 recommendations from the National Commission, aiming to overhaul the UK Medical Devices Regulations 2002.
• A major shift towards lifecycle regulation will use Predetermined Change Control Plans to manage adaptive algorithms and a new Master File system for general-purpose AI models.
• The government is committing major funding to technology and digital initiatives to make the NHS the most AI-enabled system in the world.
• New transparency measures will include a public-facing database for AI adverse incidents and mandates for tracking Unique Device Identifiers directly in electronic patient records.
• Responsibility and liability frameworks will be clarified through a dedicated working group involving the MHRA, Care Quality Commission, and NHS Resolution, rather than falling solely on frontline clinicians.
• Spring 2027 is the target for multiple consultations, secondary legislation drafts, and the publication of a detailed implementation roadmap.

Source: National Commission into the Regulation of AI in Healthcare: Government Response to the National Commission’s recommendations - Medicines and Healthcare products Regulatory Agency, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/the-44-point-plan-how-the-uk-is-rewriting-the-rules-for-healthcare-ai-712e5b

The FDA recently released a framework detailing how to use digital health technologies and AI to capture clinical trial outcomes. We explore the critical differences between verification, analytical validation, and clinical validation, and what trial sponsors must do to prove a digital measure is fit for purpose.

Key points

• Digital health technologies (DHTs) can capture continuous, real-world data outside clinical settings, creating what the FDA calls digitally derived measures (DDMs).
• Developers must clearly define the context of use and ensure the DDM captures a meaningful aspect of health, heavily involving patient and caregiver input early on.
• The FDA requires a rigorous, multi-step process: verifying the sensor works, analytically validating the algorithm measures the right metric, and clinically validating that the metric tracks the disease.
• Usability studies are critical; devices must accommodate the sensory, cognitive, and motor capabilities of the specific target population.
• Sources of error like environmental factors, incorrect device placement, or operating system updates must be continuously managed to ensure the DDM remains fit-for-purpose.

Source: Key Considerations for the Development and Use of Digitally Derived Measures for Clinical Investigations - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/digital-measures-in-trials-unpacking-the-fda-s-new-framework-2b821f

We unpack the FDA 510(k) summary for Aidoc's BriefCase-Triage, a cloud-based AI tool cleared in May 2026. Listeners will learn how it performs in flagging large and medium vessel occlusions, the real numbers behind its 1.09-minute notification time, and the specifics of its multicenter retrospective testing.

Key points

• BriefCase-Triage is a cloud-based, linux-server AI software that flags suspected large and medium vessel occlusions on CTA and mCTA images.
• The clearance expands upon a predicate device (K220709) by adding mCTA acquisition protocols to the inclusion criteria.
• In a retrospective, blinded, multicenter study, the software showed a sensitivity of 81.4% and a specificity of 85.5%.
• The positive predictive value was 38.4%, while the negative predictive value was 97.6%.
• Time-to-notification was registered at 1.09 minutes, compared to 2.23 minutes for the predicate device.
• The AI operates in parallel with the standard of care, providing low-quality, grayscale preview images meant only for prioritization, not primary diagnosis.

Source: FDA 510(k) summary K261317: BriefCase-Triage (Aidoc) - U.S. Food and Drug Administration, 2026

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-aidoc-briefcase-triage-triage-time-cut-to-1-09-minut-176e60

The European Medicines Agency has released its 2025 AI Observatory report, revealing how regulators are deploying 61 internal AI use cases to evaluate pharmaceutical submissions. Maya and Sam dive deep into the specific tools the EMA is building, the AI applications applicants are bringing to early meetings, and what this means for the future of drug development.

Key points

• The EMA is deploying tools like SCHEMA and REGULUS to independently verify safety data and process regulatory documents.
• The EU AI Act entered practical application in 2025, pushing AI from theoretical adoption to heavily governed practical use.
• Applicants are exploring radical AI use cases, from digital twins for manufacturing to predicting shelf-life without long-term stability data.
• A massive wave of EU-funded research is currently exploring AI applications in areas ranging from colorectal cancer stratification to autologous iPSC manufacturing.
• Significant regulatory gaps remain, specifically concerning model validation, explainability, and audit frameworks for regulatory-grade AI.

Source: 2025 AI Observatory report - European Medicines Agency, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/regulators-build-an-ai-arsenal-unpacking-the-ema-2025-observatory-repo-82a365

We unpack the FDA 510(k) clearance summary for Viz Subdural+, an AI tool designed to automatically label and measure collections in the subdural space from brain CT scans. Listeners will learn how the algorithm performed in retrospective testing, how it compares to its predicate device, and what its limitations mean for hospitals buying AI.

Key points

• Viz Subdural+ is cleared for the automatic labeling, visualization, and quantification of subdural collections and midline shift from non-contrast CT scans.
• The software operates in the background, analyzing images automatically and sending summary and segmentation series to a DICOM destination like a PACS.
• In a retrospective study of 203 cases, the AI achieved a mean DICE score of 73% for subdural collection volume, with a mean absolute error of 7.53.
• The testing showed significant variability, with a volume standard deviation of 13.91, highlighting why the FDA requires physician review of the outputs.
• The algorithm was cleared based on substantial equivalence to the Viz HDS predicate device, expanding capabilities from intracranial hyperdensities to subdural collections.
• The performance data is based on a limited data set from only two clinical sites, which may impact generalizability across different hospital systems.

Source: FDA 510(k) summary K250354: Viz Subdural+, Viz SUBDURAL PLUS (Viz.ai) - U.S. Food and Drug Administration, 2025

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-viz-ai-viz-subdural-quantifying-brain-scans-with-ai-59b5b6

The UK Government has released a sweeping blueprint for the future regulation of AI in healthcare, aimed at moving away from static, one-off approvals. Listeners will learn how new concepts like Predetermined Change Control Plans, function-based regulation, and Master Files for foundation models will impact medtech developers, while also unpacking the critical shift towards shared liability and mandatory AI readiness for healthcare providers.

Key points

• The current regulatory framework is viewed as restricting innovation by two-thirds of industry respondents and half of healthcare providers, prompting a shift toward proportionate, lifecycle-based regulation.
• Regulators propose an opt-in Master File system to allow developers of general-purpose foundation models to confidentially share technical data with downstream medical device makers.
• The MHRA aims to implement function-based regulation, meaning oversight will focus only on the specific medical feature of a product, rather than heavily regulating an entire general-purpose model.
• To combat the liability sink problem where clinicians shoulder disproportionate risk, the framework introduces a contract to deploy and use that explicitly allocates risk controls between manufacturers and healthcare providers.
• Post-market surveillance needs a massive overhaul, with 65% of surveyed stakeholders stating current approaches are insufficient, driving a move toward real-time monitoring and Unique Device Identifiers.

Source: National Commission into the Regulation of AI in Healthcare - UK Government, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/who-holds-the-bag-when-medical-ai-fails-the-uk-s-new-blueprint-08f65d

The FDA has granted De Novo classification to Spectral MD's DeepView AI System, a software that predicts burn wound healing. Maya and Sam unpack the device's indications, the strict special controls required, and what this new regulatory category means for future medtech developers.

Key points

• The FDA established a new Class II generic device type for software-aided adjunctive diagnostic devices used in skin wound assessment.
• DeepView AI detects deep burn tissue and highlights areas unlikely to heal within 21 days with conservative treatment.
• The device is strictly an adjunctive aid and its segmentation contour is not for definitive delineation of surgical excision borders.
• Developers must prove their AI improves diagnostic characterization compared to unaided users across diverse clinical risk factors.
• Special controls require extensive testing, including human factors, electromagnetic compatibility, and photobiological safety of light-emitting components.

Source: DeepView AI® System - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/will-ai-redefine-burn-triage-unpacking-the-fda-s-deepview-ai-de-novo-o-cc00c9

We unpack the MHRA's AI Airlock Sandbox Phase 2 Programme Report to explore how the UK regulator is pressure-testing rules for adaptive algorithms. From generative AI drifting out of scope to algorithms outperforming gold-standard pathologists, learn what happens when modern tech meets traditional regulatory frameworks.

Key points

• Generative AI can drift beyond its intended purpose over time; testing showed out-of-scope performance in approximately 39% of real-world notes without active guardrails.
• Real-world precision for AI systems can be extremely high (e.g., 0.989), which ironically reduces the effectiveness of human oversight as users become complacent.
• When AI diagnostics detect features beyond human capability, such as finding 32% more mitoses than pathologists, regulators face challenges defining the gold standard for performance evaluation.
• Pre-market testing often fails to replicate real-world conditions, making rigorous post-market surveillance essential for lifecycle management.
• Combining multiple low-risk features, like longitudinal memory and personalized biomarker tracking, can inadvertently shift a wellness app into a regulated medical device.
• Predetermined Change Control Plans offer a pathway to safely manage iterative AI updates, provided they are structurally linked to continuous real-world performance monitoring.

Source: AI Airlock Sandbox Phase 2 Programme Report - Medicines and Healthcare products Regulatory Agency, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/inside-the-mhra-ai-airlock-testing-the-boundaries-of-medical-software-def844

The FDA has granted 510(k) clearance to the Bayesian Health Sepsis Flagging Device, a continuous AI monitor that uses 115 electronic health record inputs to predict sepsis risk. Maya and Sam break down the 10-page clearance document, exploring the device's clinical validation study, its 79.4% positive percent agreement, and what an 11.7% positive predictive value means for alert fatigue.

Key points

• The FDA cleared the Bayesian Health Sepsis Flagging Device as a Class II Software as a Medical Device to aid in the early detection or risk prediction of sepsis developing within 24 hours.
• The algorithm continuously monitors 115 patient parameters from the electronic health record, including chief complaints, lab measurements, vitals, and consult orders.
• The clinical validation study included 7,732 hospital encounters across 4 distinct hospitals, utilizing a two-part adjudication process to establish truth.
• The device missed its pre-specified 80% encounter-level Positive Percent Agreement target, landing at 79.4%, but was cleared based on the totality of evidence.
• At an observed sepsis prevalence of approximately 3%, the device demonstrated a flag-level Positive Predictive Value of 11.7%.
• The clearance requires a post-market performance management plan to monitor real-world performance degradation over the device's market lifetime.

Source: Bayesian Health Sepsis Flagging Device - U.S. Food and Drug Administration, 2026

This spot is available. Reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsor the show

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/fda-clears-bayesian-s-ai-sepsis-tool-analyzing-the-data-behind-the-510-478f32