Locally hosted medical AI agents achieved competitive diagnostic accuracy in retrospective simulations, but their reliability depended on the model, dataset and configuration. Agreement across repeated encounters helped identify…

Symptoms documented before and during radiotherapy improved prediction of radiation pneumonitis in a retrospective lung cancer cohort. The strongest model combined natural language processing (NLP) features with demographic…

Maternal pregnancy records improved prediction of several infant conditions when combined with infants’ own records. MoChiAgent links these longitudinal predictions to literature-supported reports, although prospective clinical benefit remains…

A multiomics analysis identified molecular differences associated with progression within clinically low/intermediate-risk prostate cancer. Higher ZNF268 expression marked better outcomes, while a related 10-gene Rewiring Score identified patients…

Broader models, smaller hardware, richer biological signatures: this week’s papers stretch medical AI’s capabilities. The harder question is whether their gains survive the demands of clinical evaluation.

A soft ionogel that combines mechanical durability with stronger pressure-generated electrical signals offers a promising route to wearable blood pressure estimation. Liu, Gong, Zeng and colleagues demonstrated the…

A weakly supervised AI model classified lymph node whole-slide images into four categories with 92.91% accuracy in internal evaluation. Its results highlight both the potential of slide-level supervision…

Large language models can generate spine diagnoses from radiology report findings, but their performance depends on how the task is framed. Across 20,277 reports, diagnostic options supported recall,…

Routine clinical and blood data supported AI predictions of immunotherapy outcomes in advanced non-small cell lung cancer (NSCLC), and explainable AI improved some measures of physician performance. However,…

Recent Gastroenterology & Hepatology coverage shows AI moving beyond recognition: exposing occult cancer risk, redefining fibrosis, joining multimodal workflows, and testing whether apparent specialist fluency reflects real clinical…