MedAI Digest
Two Rings, One Signal: What Bilateral PPG Adds to Cardiovascular Monitoring
Recording photoplethysmography from both hands produced small but statistically significant gains over the best single-hand measurements. The clearest value was in heart-rate estimation and in using agreement between…
Reading Tumor Geography: AI Predicts Survival in Resected Pancreatic Cancer
Routine histology may contain substantially more prognostic information than conventional pathological assessment captures. In 1,020 patients with resected pancreatic ductal adenocarcinoma, HPSurv quantified spatial tumor heterogeneity and predicted…
Beyond ROC-AUC: A Clinically Focused Biomarker Framework for MASLD
A shallow neural network combining routine clinical data with liver-stiffness measurements outperformed FIB-4 across clinically important stages of fibrosis in MASLD. Just as importantly, the study reframed biomarker…
Mapping Facial Vitiligo in 3D: AI Cuts Assessment Error by Half
An AI-assisted 3D system substantially improved dermatologists’ estimation of facial vitiligo extent, reducing mean absolute error by 53.4%. By mapping segmented lesions onto reconstructed facial surfaces, the method…
Neurology & Neuroscience: From Pattern Recognition to Decision-Ready Signals
Neurology AI is moving beyond isolated accuracy claims: blood, behavior, imaging, and wearable signals are being tested for earlier detection, calibrated triage, and adaptive rehabilitation. The decisive issue…
Digital Health & Informatics: From Model Performance to Evidence-Ready Care
The current direction in Digital Health & Informatics is clear: strong model scores matter less in isolation than calibration, workflow fit, human oversight, and evidence that systems remain…
More Movement, Less Cancer Risk? What 24-Hour Accelerometry Adds
How time is distributed across the entire day may matter for cancer risk. In this UK Biobank analysis, more movement—and, to a lesser extent, more standing—was associated with…
Gated CT–Clinical Fusion Improves Hearing Prognosis After Cholesteatoma Surgery
Preoperative temporal-bone CT added meaningful predictive value when combined conservatively with routine clinical data in this retrospective cohort of cholesteatoma surgery patients. A gated fusion model achieved lower…
Geriatrics & Aging: From Disease Labels to Measurable Resilience
Recent Geriatrics & Aging coverage shows AI shifting from narrow diagnosis toward phenotyping brain aging, integrating distributed risk signals, and forecasting the independence consequences that matter most to…
Multimodal AI Matches Senior Endoscopists in EGD Diagnosis and Reporting
MLLM-EDR combined endoscopic images with expert-verified medical knowledge to diagnose gastrointestinal disease and draft EGD reports. Across 19 diagnoses, it exceeded vision-only AI baselines and junior endoscopists, with…
Can Reinforcement Learning Track Coagulation Risk Over Time?
Coagulopathy in critical illness is not static: bleeding and thrombosis risks can coexist and change as laboratory results, treatments, and time evolve. Wang, Kang, Zhu et al. present…
Can a Few Words from an EMS Call Predict Lifesaving Care?
Brief, routine free-text chief complaints contained enough signal to predict many out-of-hospital lifesaving interventions. In a nationwide temporal validation study, lexical machine-learning models achieved good-to-excellent discrimination for six…
A Transparent CXR Foundation Model Maps Radiographic Features to Broad Disease Risk
A contrastively pretrained chest-radiograph foundation model extracted clinically meaningful signals from a single CXR across a broad range of present and future disease phenotypes. Crucially, the study connects…
A Double Neural Bypass Restores Assisted Hand Use and Touch After Complete Tetraplegia
A first-in-human “double neural bypass” combined brain-controlled hand assistance with stimulation-based rehabilitation in a person with chronic complete tetraplegia. The system supported functional grasping and tactile feedback while…
Can AI Replace Contrast for Chronic MI Scar Imaging? A Multicenter Test of Virtual Native Enhancement
Artificial intelligence–generated virtual native enhancement (VNE) may allow chronic myocardial infarction scar to be assessed without gadolinium in many patients. In this prospective multicenter study, confident VNE interpretations…
PM2.5 Composition Matters: Organic Carbon Emerges as a Strong CHD Signal in China
In this nationwide Chinese cohort, the cardiovascular burden of PM2.5 was not uniform across its chemical components. Long-term exposure to organic carbon showed the strongest and most consistent…
Lewy Body Co-Pathology Is Linked to Faster Brain Aging in Alzheimer’s Disease
Lewy body co-pathology may mark a more aggressive neurodegenerative course in Alzheimer’s disease. In this cohort study, participants with biomarker-defined Alzheimer’s disease and Lewy body pathology had the…
A 13-Feature Blood Panel Predicts Near-Term Cognitive Conversion in Down Syndrome
Blood biomarkers spanning Alzheimer-related neurodegeneration, inflammation, and vascular biology helped predict 16-month conversion from cognitive stability to MCI or Alzheimer’s disease in adults with Down syndrome. In the…
One Compact AI Model Takes on Ten Gait Assessment Tasks
Wearable inertial sensors can support more than a single gait classification task. In this computational study, a hierarchical multi-task model jointly assessed diagnostic, severity, demographic, and functional outcomes…
Eight Routine Measures Predict 10-Year Cognitive Impairment Risk in Rural China
A random-forest model built from eight routinely available indicators moderately predicted 10-year incident cognitive impairment in a rural Chinese cohort. Its value is as an interpretable triage aid…
