The Wire: 26 October 2026

The ten stories

1. Samsung Health Expands AI-Powered Diabetes Prevention and Chronic Condition Support for Older Adults Through Lark Health Samsung (company announcement) · 26 Oct

   Eligible Samsung Health users 65 and over get Lark's AI-powered Medicare Diabetes Prevention Program first, with cardiometabolic and behavioral-health programs to follow under the CMS ACCESS model, pairing AI coaching with clinical oversight at no extra out-of-pocket cost.

2. Radiomics-GenAI: a novel model for laryngeal lesion classification BMC Medical Imaging · 24 Oct

   Generative augmentation of the radiomic feature manifold - without generating synthetic images - improved laryngeal lesion classifiers on an unseen test set, with Radiomics-GAN reaching a median F1 of 0.851 versus 0.806 for a SMOTE-balanced baseline.

3. Examining reliance patterns on AI advice in medical imaging: a mixed-methods, randomised, crossover experiment The Lancet Digital Health · 23 Oct

   In a crossover experiment on CT intracranial-haemorrhage review, under-reliance on correct AI advice harmed diagnostic performance more than over-reliance in absolute terms; correct explainable-AI advice improved performance (OR 1.84) and cut under-reliance, but did not cut reading time.

4. Human oversight of erroneous clinical AI and decision support: a systematic review and meta analysis npj Digital Medicine · 23 Oct

   Across 46 reports, clinicians were far more often correct with correct than incorrect automated advice (pooled OR 10.19); in four real-world ECG studies 26.6% of erroneous AI reads were never corrected. The authors argue oversight should be measured as an error-interception capability, not assumed.

5. The unannounced guest: The challenge of client-supplied AI in teleradiology AuntMinnie Europe (ESER Committee on Teleradiology) · 22 Oct

   The European Society of Emergency Radiology's teleradiology committee describes radiologists opening worklist studies to find client-side AI outputs - heat maps, markers, generated findings - they never chose and may not recognize, and sets out the governance and liability questions that follow.

6. Eko Health announces FDA clearance and launch of CORE 300 stethoscope Eko Health (company announcement via PRNewswire) · 22 Oct

   The company says CORE 300, cleared for pediatric and adult patients, is the first stethoscope to combine traditional analog and digital listening in a single chestpiece, delivering FDA-cleared AI structural-murmur detection at a price in line with premium traditional stethoscopes.

7. Ambient voice technology rolled out across Mid Cheshire Mid Cheshire Hospitals NHS Foundation Trust (official) · 22 Oct

   After a months-long trial in SDEC, dermatology, rheumatology and urology, Lyrebird AVT is rolling out across the trust's hospitals and community services, including ED, paediatrics, orthopaedics, stroke and the Minor Injuries Unit, generating notes from clinical conversations in real time.

8. WHO Director-General's remarks at the signing of the Seoul Declaration during the fifth World Bio Summit WHO (primary) · 21 Oct

   Signing the first Seoul Declaration on AI in Healthcare with South Korea, the WHO Director-General said real-world evidence remains limited, a tool should be shown to work where it will be used before scaling, oversight must be proportionate to risk, and health workers must remain accountable.

9. Agentic Automation of BT-RADS Scoring Using an End-to-End System for Standardized Glioma Follow-up Assessment Radiology: Artificial Intelligence · 21 Oct

   A multiagent LLM-plus-CNN system that extracts clinical variables from notes and applies BT-RADS decision logic with volumetrics scored 76.0% accuracy on 492 evaluable posttreatment glioma MRI exams versus 57.5% for the initial assessments (+18.5 points, P<.001) against expert reference.

10. TRAM-AI: A Framework for Transparency in Reporting of Medical Imaging Datasets for Artificial Intelligence Development Radiology: Artificial Intelligence · 21 Oct

    A modified-Delphi panel of 33 international experts produced a 14-item checklist for verifying whether medical imaging datasets billed as publicly available are functionally usable for AI development, pilot-validated on 10 public datasets.