HisToSpatialCNV turns routine H&E morphology into spatial predictions of genomic gain and loss. Across breast and skin cancer datasets, the interpretable deep-learning framework recovered clinically and biologically relevant…

Recent coverage shifts the field’s center of gravity from benchmark performance to conditional utility: who benefits, which shortcuts models exploit, how biology is encoded, and whether routine signals…

An electronic health record–integrated large language model identified surgical patients potentially needing hospitalist comanagement with high sensitivity and a 0.97 negative predictive value. The findings support its use…

Paramagnetic rim lesions identified people with multiple sclerosis at increased risk of future disability accumulation, particularly progression independent of relapse activity. A greater lesion burden also predicted faster…

Digital health AI is becoming more specialized, multimodal, and clinically legible. The harder challenge now spans the full lifecycle: preserving privacy and fairness, supporting human interpretation, and responding…

A shoe-mounted system powered by biomechanical energy classified four locomotor activities with 95.4% accuracy while performing inference locally. Its complete power demand was only 108 μW, low enough…

A radiography-based AI tool showed broadly consistent fracture-detection performance across three European hospitals. When clinicians used it, pooled sensitivity increased by 10.6 percentage points without a significant change…

An AI-derived prostate MRI quality score closely tracked clinically significant cancer detection. As the lowest-quality examinations were progressively excluded, performance improved for both a cancer-detection model and routine…

A hospital-trained AI-ECG model lost substantial discriminatory power when deployed among community-dwelling older adults. PREVUE-VALVE shows that diagnostic AI performance depends not only on disease prevalence, but also…

LiON combined high diagnostic accuracy with measurable workflow benefits in liver CT interpretation. In a prospective trial of 10,333 patients, AI–human review confirmed 51 previously overlooked lesions, including…