Recent coverage shows the field shifting from isolated prediction toward context-aware imaging, calibrated risk stratification, and LLM systems designed around clinician interaction, workflow, and appropriate reliance.

Recent Gastroenterology & Hepatology coverage shows AI shifting from finding overlooked disease to supporting consequential choices in liver care, endoscopy, and biologically defined patient stratification.

In a multicenter randomized trial, Retina4IRD substantially improved specialists’ ability to predict clinically relevant genotype categories before genetic testing. Top-5 accuracy rose from 67.3% with specialist assessment alone…

Oncology AI is broadening from detection to screening, surgery and molecular triage—but this batch makes the standard clear: multicentre validation, realistic prevalence and rigorous benchmarks matter as much…

In the 1460s, whether a village invested in vineyards was strongly associated with its Greek or Albanian identity. By the 1580s, that cultural divide had narrowed sharply as…

Open-weight large language models can identify many people meeting clinical high-risk criteria for psychosis and approximate researcher symptom ratings—but they are not ready to diagnose independently. Their strongest…

A review of 4,952 second-level Medicare Part D appeals found that fewer than one in five denials of anticancer medication claims were overturned. More than half of unfavorable…

Surgical AI coverage is moving upstream: toward tools that shape referrals, operative choices, recovery planning and organ preservation—while benchmarks, computable outcomes and reporting quality make those decisions more…

CLEAR combines strong chest X-ray classification with an unusually direct form of interpretability: every prediction can be decomposed into clinically meaningful radiological observations. In external and supervised evaluations,…

General-purpose large language models classified standardized health checkup results with high accuracy and reproducibility when given carefully designed prompts. Yet performance varied by model, examination item, sex, and…