The Wire: 23 October 2026

The ten stories

1. New AI tools could help spot prostate cancer more accurately NICE (primary) · 23 Oct

   NICE draft guidance recommends five AI technologies as options to help detect prostate cancer on MRI in the NHS while further evidence is generated, with potential to improve the accuracy and consistency of prostate MRI interpretation, particularly where specialist expertise is scarce.

2. FDA grants 510(k) clearance for BoneEnhance, first ML layer on RIVANNA's Accuro XV platform PRNewswire (company PR, attributed) · 22 Oct

   The module segments bone and surrounding tissue and renders 3D skeletal anatomy on the Accuro XV ultrasound system, which was cleared for B-mode MSK imaging in April 2026; an injury-detection module is next on the roadmap.

3. Sequential diagnosis with language models Science · 22 Oct

   Microsoft's SDBench turns 304 NEJM cases and 325 MIMIC-IV emergency encounters into interactive diagnosis; a stepwise AI panel hit 83.2% and 80.3% accuracy versus 36.1% and 46.8% for generalist physicians, and a budget-capped setting still beat them on accuracy and test cost (80.9%, 73.2%).

4. FDA Clears AI-Enabled FFR-IQ Software for Estimating Fractional Flow Reserve from CCTA Diagnostic Imaging · 21 Oct

   Elucid's FFR-IQ won 510(k) clearance to estimate fractional flow reserve non-invasively from coronary CT angiography, folding plaque composition and anatomy into the physiology estimate; the company says it may cut unnecessary catheterization procedures.

5. HRDPath: an explainable multi-model deep learning architecture for predicting homologous recombination deficiency from histopathology images Nature Communications · 21 Oct

   HRDPath predicts homologous recombination deficiency - the biomarker guiding PARP-inhibitor treatment in ovarian cancer - directly from whole slide images, reaching AUC 0.846 across three cohorts in this research study, with an interpretability pipeline aimed at trust.

6. Interpretable Deep Learning for Opportunistic Screening of Left Ventricular Dysfunction Using Chest Radiographs Radiology: Artificial Intelligence · 21 Oct

   A deep learning model flagged reduced ejection fraction (LVEF 40% or less) on plain chest X-rays with AUCs of 0.92-0.97 in health checkup cohorts, and prescreening in those cohorts could have cut echocardiography volume by up to 85% while identifying every low-EF case.

7. Paradoxical inflation of prediction model performance measures in minority patient subgroups: the role of upstream bias npj Health Systems · 21 Oct

   A model predicting persistent postoperative opioid use scored higher in Black patients, who made up under 5% of the sample, than in White patients - but subgroup matching attenuated the gap, showing upstream biases can inflate apparent performance in minoritized groups and distort fairness audits.

8. Thousands of NHS Wales staff building AI skills through national Microsoft Copilot campaign Digital Health and Care Wales (official) · 21 Oct

   A national campaign by the NHS Wales Microsoft 365 Centre of Excellence is training thousands of staff to use AI in everyday work, building confidence and skills around the Copilot rollout.

9. Expertise modulates automation bias and sentinel behavior in human-AI collaborative diagnosis of neonatal pneumoperitoneum npj Digital Medicine · 20 Oct

   In a 1,750-interpretation crossover study, trainees accepted 52.0% of an AI's deliberately injected errors while expert neonatologists overrode 91.7%, showing high AI reliability can induce automation bias in juniors; the authors released an open-source training sandbox to inoculate clinicians.

10. SimBioSys Secures FDA Clearance for TumorSight Risk, Introducing the World's First AI-Powered, MRI-Derived Breast Cancer Recurrence Risk Assessment SimBioSys (company announcement) · 15 Oct

    The company says newly 510(k)-cleared TumorSight Risk is the first AI-powered, MRI-derived prognostic tool for breast cancer recurrence risk, producing five- and ten-year estimates for HR-positive, HER2-negative early-stage disease; validation covered 2,129 patients at four institutions.