PEAL offers a one-shot federated learning approach that reproduces pooled individual-level linear mixed-model results while keeping patient records local.

ARTIMES proposes a more reproducible way to measure pleural mesothelioma response on CT, replacing difficult linear measurements with AI-assisted tumour volume assessment.

A new JAMA Otolaryngology study reports that preoperative AI models can stratify short-term complication risk after tongue cancer surgery using routinely available clinical variables.

Can a diagnostic AI become fairer by being guided to look at the lesion itself? Zhu, Liu, Yuan et al. show that attention alignment can reduce gender disparity…

Software is no longer peripheral to medical devices. A two-decade analysis of FDA authorizations shows that digital functionality has become a substantial and growing feature of devices entering…

A real-world pilot study suggests that LLM-orchestrated PET/CT agents can automate the technical pipeline from raw DICOM data to draft report, but not replace physician judgment.

A multicenter study in NPJ Digital Medicine reports that TriageMaster-70B can triage possible acute coronary syndrome early, using only patient narratives and vital signs before ECG or biomarker…

A web-based deep learning platform may make AS-OCT assessment of infectious keratitis more quantitative and reproducible.

Machine learning applied to the HUNT population cohorts suggests that migraine can be diagnosed from broader clinical signals and may comprise distinct data-driven phenotypes.

An autonomous medical AI agent, MIRA, was tested in a sandboxed emergency department EHR and achieved high diagnostic accuracy with strong safety signals.