A new transformer-based foundation model trained on over 500,000 UK Biobank individuals reveals hidden disease subtypes and demonstrates superior disease prediction, enabling precision medicine approaches at scale. Original paper: A foundational model encodes deep phenotyping data and enables diverse downstream applications. — NPJ digital medicine. 10.1038/s41746-026-02736-w 📄 Read the article

Researchers developed an interpretable machine learning model predicting chronic osteomyelitis recurrence following Masquelet technique surgery, achieving 91% accuracy in external validation. Original paper: MRI-based radiomics explainable model for predicting recurrence of limb chronic osteomyelitis in limb bones treated by Masquelet technique. — Bone & joint research. 📄 Read the article

A state-of-the-art large language model has demonstrated superior clinical reasoning compared to experienced physicians across multiple established benchmarks, suggesting AI systems may be ready for real-world integration. Original paper: Performance of a large language model on the reasoning tasks of a physician. — Science (New York, N.Y.). 10.1126/science.adz4433 📄 Read the article

A novel mathematical framework based on rotational symmetry transforms diffusion MRI into a universal diagnostic tool, achieving 30% better disease detection with acquisition times of just 1–2 minutes. Original paper: Geometry of the cumulant series in diffusion MRI. — Nature Communications. 10.1038/s41467-026-70018-w 📄 Read the article

A prospective study of 93,000 participants demonstrates that genetic susceptibility to diabetic retinopathy can be substantially mitigated through lifestyle modifications, even among genetically high-risk individuals. Original paper: Decoding Gene-Lifestyle Synergy in Diabetic Retinopathy Development: A Transethnic Prospective Cohort Study. — Ophthalmology. 10.1016/j.ophtha.2026.04.017 📄 Read the article

A new plasma biomarker panel combined with AI can accurately diagnose multiple neurodegenerative diseases and detect mixed pathologies, offering a non-invasive tool for precision diagnosis. Original paper: GPND-AI NULISA: A 15-Protein AI classifier for diagnosis and co-pathology profiling across neurodegenerative diseases. — Alzheimer’s & dementia : the journal of the Alzheimer’s Association. 10.1002/alz.71420 📄 Read the article

Researchers have developed COCA, a deep learning system that detects colorectal cancer from routine non-contrast CT scans with greater accuracy than experienced radiologists. This could enable cost-effective opportunistic screening of millions of existing scans annually. Original paper: Colorectal cancer detection using non-contrast CT and deep learning: a multicenter and international cohort study. — Annals of oncology : official journal of the European Society for Medical Oncology. 10.1016/j.annonc.2026.04.009 📄 Read the article

Researchers developed a deep learning algorithm using routine electronic health records to identify young adults at high risk of type 2 diabetes, dramatically improving risk stratification. Original paper: Detection of young-onset type 2 diabetes using deep learning across primary and secondary care: a nationwide, retrospective cohort study. — The Lancet. Digital health. 10.1016/j.landig.2025.100968 📄 Read the article

BraTS-PEDs fills a critical gap by providing the first large-scale, multi-institutional, open-access benchmark dataset for developing AI algorithms to segment pediatric high-grade gliomas. Original paper: Multi-Institutional Annotated Multiparametric MRI Dataset of Pediatric High-Grade Gliomas. — Radiology. Artificial intelligence. 10.1148/ryai.250902 📄 Read the article

A new automated AI framework called REDMOD can detect pancreatic cancer before it becomes visible on standard CT scans, dramatically outperforming expert radiologists at identifying the disease in its earliest, most treatable stages. Original paper: Next-generation AI for visually occult pancreatic cancer detection in a low-prevalence setting with longitudinal stability and multi-institutional generalisability. — Gut. 10.1136/gutjnl-2025-337266 📄 Read the article