AI in Medicine - curated summaries making complex issues easy to understand

Mike Rawson
  • 0.0 (0)
  • Medicine
  • Weekly

AI in Medicine - Smart Summaries is a daily podcast for clinicians, health-system leaders and medtech, pharma and digital-health teams. Each episode unpacks one important public document on artificial intelligence in healthcare, such as FDA, EMA, MHRA, WHO or NHS guidance, OECD and World Economic Forum reports, or an open-access peer-reviewed review, in a clear 15 to 20 minute conversation. Topics include clinical AI and decision support, medical imaging and radiology AI, large language models (LLMs) and generative AI in clinical practice, AI-enabled medical devices and their regulation, the EU AI Act, drug discovery and clinical trials, digital health and wearables, health-system adoption, patient safety, bias and liability. The source of every episode is linked in the show notes, and full transcripts are at ai-in-medicine-podcast.vercel.app. The hosts' voices are AI-generated. For information only, not medical advice.

  1. 10h ago

    Company Spotlight: HistoSonics Planning Tool - AI Mapping for Histotripsy

    We unpack the FDA 510(k) summary for the HistoSonics Planning Tool, cleared in May 2026. This episode explores how the standalone software uses machine learning to segment organs and map acoustic pathways for histotripsy procedures. We dive into the training data, the validation subsets, and what this clearance means for clinical workflows. Key points The FDA cleared the HistoSonics Planning Tool as a Class II medical image management and processing system based on substantial equivalence to a 2019 microwave ablation visualization device.The software uses static machine learning models to automatically segment the liver, kidneys, lungs, and ribs in CT and MRI scans to help plan acoustic pathways.Performance testing showed the liver segmentation model achieved a mean Dice score of 93%, exceeding the target of 70% or higher in 70% of cases.The validation relied entirely on retrospective datasets, including subsets of 10 patients from larger public databases, with some imaging data dating back to the period from 1991 to 1995.Because it is a planning tool, the device was cleared through software testing and did not require new animal or human clinical studies. Source: FDA 510(k) summary K252947: HistoSonics® Planning Tool (HistoSonics) - 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. Company: HistoSonics (histosonics.com) · Product: HistoSonics® Planning Tool · FDA 510(k) K252947 Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-histosonics-planning-tool-ai-mapping-for-histotripsy-752a1c

  2. 18h ago

    Why Generative AI Trials Fail the Reproducibility Test

    A 2026 systematic review in the Journal of medical Internet research reveals critical reporting gaps in randomized controlled trials of generative AI. Maya and Sam unpack why current reporting standards fail to capture the mechanics of large language models, leaving major blind spots in model versioning, prompt engineering, and input data quality. Key points A review of 74 trial reports evaluating generative AI found that 0% detailed how they handled poor-quality or unavailable input data.Only 23.0% of reports specified the exact version of the AI algorithm used, while just 27.0% explicitly reported the system prompt.Current reporting standards like CONSORT-AI were built for conventional, locked AI models and suffer from structural mismatches when applied to conversational AI.Trials where laypeople were the intended users reported how outputs contributed to decision-making at a drastically lower rate (15.8%) than clinician-facing trials (100%).Failing to report inference settings, system prompts, and access modalities makes reproducing generative AI trials virtually impossible for healthcare buyers. Source: Reporting Gaps and Challenges in Randomized Trials of Generative AI for Medical Use: Insights From a 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. Authors: Jae Eun Ko, Ji Woo Hwang, Sung Min Kim, Ji Yean Kwon · DOI: 10.2196/94543 Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/why-generative-ai-trials-fail-the-reproducibility-test-828fd6

  3. 1d ago

    Company Spotlight: Iterative Health SKOUT - Live polyp detection with 0ms delay

    This episode examines the FDA 510(k) summary for Iterative Health's SKOUT system, a computer-aided detection tool that places a blue bounding box around potential colorectal polyps in real time. We unpack the technical specs, the zero-millisecond video delay, the predicate device comparison, and what key performance data is missing from the public clearance document. Key points The SKOUT system is cleared as a Class II software device to assist gastroenterologists by highlighting suspected colorectal polyps with a two-dimensional blue rectangular outline during white light colonoscopies.The FDA clearance (K253664) is based on substantial equivalence to an earlier version of the SKOUT system (K251126), not on a new standalone clinical trial.The software features smart pausing, disabling the polyp detection marker when an endoscopic surgical tool enters the frame or if lighting is inadequate.Bench testing reported a video delay due to the device and marker annotation of exactly 0ms, with a minimum resolution of 1.1ms.The updated software was retrained for compatibility with Endocuff devices, multiple polyps, and sessile serrated lesions, aiming for a reduced false positive rate.While the summary notes that performance testing evaluated true positives and false positives on an expanded dataset, it does not disclose the actual dataset size or the accuracy percentages. Source: FDA 510(k) summary K253664: SKOUT system (Iterative 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. Company: Iterative Health (iterativehealth.com) · Product: SKOUT system · FDA 510(k) K253664 Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-iterative-health-skout-live-polyp-detection-with-0ms-522c3b

  4. 1d ago

    Who Pays When Clinical AI Fails? The AMA's Blueprint for Medical Malpractice

    The American Medical Association's October 2026 issue brief tackles the medical community's biggest fear regarding artificial intelligence: liability. This episode explores the AMA's push to align legal accountability with actual control, shifting the burden away from individual physicians and onto the developers who build clinical AI and the health systems that deploy it. Key points The AMA argues that liability for AI-related harm should rest with the parties best positioned to know the risks and prevent them, rather than defaulting to the physician.Physicians should not be held responsible for errors they cannot reasonably detect, such as upstream data errors or hidden algorithms causing triage delays.Developers cannot shift responsibility to physicians simply through warnings; they remain accountable for design, validation, and system architecture.Deploying health organizations bear an institutional duty of care for selecting AI tools, validating their local performance, and monitoring ongoing safety. Source: Issue Brief: Liability for Clinical AI - 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/who-pays-when-clinical-ai-fails-the-ama-s-blueprint-for-medical-malpra-03a409

  5. 2d ago

    The Competency Cliff: Why Hospitals Must Turn Off AI's Continuous Learning

    IQVIA's white paper outlines a strategic framework for responsible AI governance in regulated healthcare systems. The hosts unpack why disabling self-learning features is often necessary for regulatory compliance, how automation creates a dangerous 'competency cliff' for junior staff, and why AI literacy is now a strict legal mandate rather than an optional skill. Key points The core requirement for healthcare AI is that 'Human-in-the-loop is non-negotiable' because clinical and ethical judgment cannot be automated.Hospitals face a 'competency cliff' as AI automates the routine tasks that traditionally provided the training ground for junior clinicians and quality professionals.Under current global regulations, validating a self-learning algorithm often requires turning the continuous learning feature off to maintain a fixed, auditable model.The EU AI Act and FDA guidance have transformed AI literacy from a competitive advantage into a strict regulatory compliance mandate.AI's success metrics must shift away from traditional headcount reduction toward patient-centric outcomes like early disease detection and safety improvements.Effective governance requires multidisciplinary teams—business, quality, and technology experts—working together to review and monitor applications before deployment. Source: The Human Edge of AI: Building Trust, Safety and Value in Regulated Healthcare Systems - 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/the-competency-cliff-why-hospitals-must-turn-off-ai-s-continuous-learn-523c03

  6. 2d ago

    Company Spotlight: Qure.ai qER-CTA - AI Stroke Triage and the 6.36-Minute Notification Race

    We unpack the FDA's 510(k) summary for Qure.ai's qER-CTA, an AI triage tool designed to detect large vessel occlusions on head CT scans. We explore its clinical performance data, how it compares to its predicate device Viz LVO, and the strict workflow limitations hospitals need to know before buying. Key points qER-CTA uses deep learning to analyze CT angiograms for suspected large vessel occlusions in the ICA and M1 segment of the MCA.In a standalone study of 584 scans, the software demonstrated 91.35% sensitivity and 91.86% specificity, with a mean notification time of 6.36 minutes.The FDA cleared it based on substantial equivalence to Viz LVO, placing it in the highly competitive Radiological Computer Aided Triage and Notification Software category.The system generates a passive notification flag in the worklist rather than proactively alerting users or altering the worklist order itself.Images viewed on the mobile application are compressed and strictly for informational purposes; clinical decisions require reviewing non-compressed images on a diagnostic viewer. Source: FDA 510(k) summary K251610: qER-CTA (v1.0) (Qure.ai Technologies) - 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. Company: Qure.ai Technologies · Product: qER-CTA (v1.0) · FDA 510(k) K251610 Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-qure-ai-qer-cta-ai-stroke-triage-and-the-6-36-minute-082141

  7. 3d ago

    Company Spotlight: Therapixel MammoScreen 5 - AI Breast Cancer Screening Clearance

    This episode unpacks the FDA 510(k) summary for Therapixel's MammoScreen 5, an AI tool cleared to assist radiologists in reading mammograms. Listeners will learn how the software categorizes and scores lesions, the specific performance metrics across different image combinations, and the notable gaps in the demographic data used for its standalone performance testing. Key points MammoScreen 5 is a software-only device cleared by the FDA as a concurrent reading and reporting aid for screening mammograms.The AI outputs a MammoScreen Score from one to ten and classifies findings as mass/asymmetry, distortion, or calcifications.Clearance was based on substantial equivalence to MammoScreen 4, using standalone performance testing on 9,789 patients.The testing dataset contained significant demographic reporting gaps, with 9,461 patients listed as not reported for race and ethnicity.The software demonstrated non-inferiority across new scenarios, including supplementary views and specific combinations of current and prior imaging.The FDA explicitly warns that patient management decisions should not be made solely based on the software's analysis. Source: FDA 510(k) summary K260714: MammoScreen® (5) (Therapixel) - 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. Company: Therapixel (therapixel.com) · Product: MammoScreen® (5) · FDA 510(k) K260714 Full transcript: https://ai-in-medicine-podcast.vercel.app/episodes/company-spotlight-therapixel-mammoscreen-5-ai-breast-cancer-screening-5aeaf6

  8. 3d ago

    Closing the App Loophole: Australia's TGA Rethinks Digital Mental Health Regulation

    Australia's Therapeutic Goods Administration is consulting on whether to close a regulatory loophole that allows certain digital mental health tools to bypass medical device oversight. We explore the proposed options, the risks of unregulated mental health software, and what a shift toward stricter compliance means for developers and clinicians worldwide. Key points Since 2021, certain digital mental health tools have been excluded from TGA regulation if they link to established clinical practice guidelines, but a 2024 review suggests this is no longer fit for purpose.The TGA argues the current exclusion criteria fail to account for product intended purpose, user groups, clinical claims, or the use of complex technologies like artificial intelligence.The consultation outlines 3 options: keep the current exclusion, remove it entirely so tools become regulated medical devices, or modify the criteria to only exclude very low-risk tools.If the exclusion is removed or modified, affected software sponsors would face a 2-year transition period to comply with requirements like ARTG inclusion and adverse event reporting.The TGA is aligning its perspective with international regulators like the US FDA and the UK MHRA, which generally regulate digital mental health tools unless they are strictly for general wellness or possess low functionality. Source: Consultation: Reviewing the exclusion for certain digital mental health tools - Therapeutic Goods 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/closing-the-app-loophole-australia-s-tga-rethinks-digital-mental-healt-d0aaaf

See All (145)

About

AI in Medicine - Smart Summaries is a daily podcast for clinicians, health-system leaders and medtech, pharma and digital-health teams. Each episode unpacks one important public document on artificial intelligence in healthcare, such as FDA, EMA, MHRA, WHO or NHS guidance, OECD and World Economic Forum reports, or an open-access peer-reviewed review, in a clear 15 to 20 minute conversation. Topics include clinical AI and decision support, medical imaging and radiology AI, large language models (LLMs) and generative AI in clinical practice, AI-enabled medical devices and their regulation, the EU AI Act, drug discovery and clinical trials, digital health and wearables, health-system adoption, patient safety, bias and liability. The source of every episode is linked in the show notes, and full transcripts are at ai-in-medicine-podcast.vercel.app. The hosts' voices are AI-generated. For information only, not medical advice.

Information