PRISM is a large MRI-specific foundation model designed to separate anatomy from sequence-related contrast. Across 44 downstream tasks, it ranked first in 39, indicating that broad multi-sequence pretraining…

Target trial emulation improved the structure of observational analyses, but it did not recover randomized-trial mortality effects for two widely studied heart-failure treatments. The findings underscore how strongly…

This week’s strongest papers share a hard lesson: medical AI earns its place not by seeing more data, but by making reliable decisions amid missing inputs, noisy signals,…

Recent Nephrology & Urology coverage shows the field’s center of gravity shifting: multimodal performance matters, but portability, alert burden, workflow fit, and decision consequences now determine whether AI…

Clin-FFA-VLM was designed to interpret fundus fluorescein angiography (FFA) in the same sequence used by retina specialists: recognize lesions, produce a report, then make a diagnostic decision. In…

Digital Health & Informatics is moving beyond isolated accuracy claims: the current pattern is context-aware adaptation, interpretable decision support, and tougher evidence for whether sensing and prediction can…

Strong discrimination did not guarantee stable operational performance. In a six-year simulation of emergency-triage admission prediction, substantial covariate drift accompanied worsening calibration and bed-demand error, while periodic retraining…

A Vision Transformer model trained on static B-mode ultrasound images predicted air-enema reduction failure in children with ileocolic intussusception with high accuracy in prospective validation. Its performance exceeded…

Longitudinal patterns in diagnoses, prescriptions, and procedures may identify people at risk of recorded Alzheimer’s disease and related dementias (ADRD) years before diagnosis. In a large retrospective study,…

Frontier large language models achieved strong scores on cardiology residency in-service examination items, reaching 86.4% accuracy. Yet this study shows why an examination score alone is an incomplete…