Machine learning models for osteosarcoma often fail when applied to different hospitals. A multi-registry approach dramatically improves their generalizability.

High triglycerides don't reliably predict who develops severe pancreatitis. A novel graph neural network called DeepLipiDecipher reveals why: it's not the quantity of lipids, but their chemical structure…

Researchers developed a deep learning model that predicts lobar-specific emphysema progression from baseline CT scans, significantly outperforming traditional clinical approaches.

AI triage can safely reduce radiologist workload by 77% in mammogram screening while maintaining cancer detection rates.

A striking paradox in AI-assisted medical imaging: clinicians who need help most benefit most from large language models, while expert radiologists gain surprisingly little.

Surgical randomized controlled trial abstracts are notoriously incomplete, but a tightly constrained large language model pipeline using GPT-4o can substantially improve CONSORT reporting at scale.

A deep learning model predicts pathological response across multiple neoadjuvant regimens, enabling clinicians to personalize treatment intensity for locally advanced rectal cancer.

Virtual simulation systems are increasingly important in dental education, with emerging AI opportunities. A systematic review of 57 studies reveals the current landscape and future potential of AI-integrated…

Medical decisions often reveal substantial expert disagreement. A kidney biopsy algorithm improves performance by modeling diverse expert patterns rather than enforcing consensus—a paradigm shift toward embracing clinical variability.

PolyPulse, an AI-enabled radar, measures blood pressure and pulse transit time contactlessly with clinical accuracy, eliminating the need for wearables or contact sensors.