Ophthalmic AI is moving beyond image classification toward generated angiography, multimodal triage, longitudinal measurement, molecular risk, therapeutic design, and a reckoning with AI’s role in the literature itself.
Multimodal artificial intelligence models predicted long-term mortality after transcatheter aortic valve implantation more accurately than established surgical risk scores. Crucially, their performance held in an external Japanese cohort…
HealthFlow brings planning, execution, quality control, and learning into a single framework for automated electronic health record analysis. Across five benchmarks, it achieved the highest reported performance while…
A retina-specific foundation model did not universally outperform conventional deep-learning models. RETFound’s clearest advantage emerged in low-data prediction of systemic diseases, while abundant ocular labels largely erased the…
Radiology AI is moving beyond classification toward grounded interpretation, predictive biomarkers, synthetic imaging, and institution-ready learning. The central challenge is shifting from benchmark breadth to reliable clinical use.
A deep learning model distinguished surgical wound photographs suggestive of incisional surgical site infection with strong internal and external discrimination. Its performance also exposed a crucial deployment issue:…
A machine-learning analysis separated amyloid-enriched functional brain-network alterations from variation shared across older adults. The resulting signatures predicted cerebrospinal fluid biomarkers and mapped onto distinct cognitive, molecular, and…
How older adults complete routine questionnaires may provide useful signals of cognitive impairment. Across five countries, a jointly trained AI model identified these signals with promising accuracy and…
Using nationwide registry data from more than 3.5 million adults, FRACTURE-ML predicted hip fracture risk up to 10 years without an in-person assessment. A practical 35-variable version detected…
Critical-care AI is moving beyond isolated prediction toward adaptive treatment support, interpretable physiologic surveillance, and workflow-integrated action. The decisive test is now whether these systems improve care, not…
