MedAI Digest

HOPE, a transformer-based AI model, predicts hyperprogressive disease risk in advanced liver cancer patients on PD-1 inhibitors by combining CT imaging with clinical factors.

Familiar names may unlock a window into the brains of comatose patients. A new study shows EEG responses to patients' own names can reliably predict recovery outcomes in…

A deep learning approach reveals the genetic architecture of brain white matter with unprecedented detail, uncovering thousands of previously unknown genetic variants and potential drug targets.

Medical imaging AI models degrade in performance after deployment. ReclAIm—a multi-agent framework—automates detection and correction of this decline while preventing catastrophic forgetting.

AI foundation models support less experienced clinicians but expert dermatologists outperform all algorithms in skin lesion diagnosis.

A landmark randomized controlled trial demonstrates that electroacupuncture provides clinically meaningful symptom relief in knee osteoarthritis, with exploratory evidence suggesting potential anti-inflammatory and cartilage-protective effects.

A mixed-methods study of 930 Chinese medical students reveals significant GenAI dependence linked to reduced critical thinking, raising urgent questions about responsible integration in medical education.

A systematic review of large language models in clinical practice finds promising early results but reveals alarming gaps in evidence quality and reporting standards that must be addressed…

A prospective study reveals that freely available AI models significantly underperform dermatologists in diagnosing nail disorders, raising important questions about AI's role in clinical dermatology.

Recent Anesthesiology & Critical Care coverage reveals a field in transition: static risk calculators are giving way to dynamic trajectory models that predict deterioration hours ahead, reinforced by…

Researchers developed MCEN, a Mamba-based deep learning model that accurately predicts chemotherapy response in breast cancer using pre-treatment needle biopsies, offering a clinical decision-support tool for personalized treatment.

SKELEX demonstrates how foundation models solve a critical bottleneck in medical AI: the need to retrain from scratch for each new disease or anatomical region. This self-supervised approach…

Contactless smart home sensors combined with AI can identify individuals at risk of cerebrovascular disease by detecting subtle behavioral changes, potentially enabling earlier intervention.

Researchers developed a proteomics-based machine learning model that predicts diabetic retinal neurodegeneration from plasma proteins, enabling early risk stratification via a simple blood test.

A new study shows that AI-based image reconstruction for breast tomosynthesis maintains cancer detection rates while improving specificity and reducing the number of images radiologists must interpret.

An AI framework automatically analyzes cardiac MRI to identify patterns that better predict major complications after heart attacks than conventional methods.

Psychiatry's recent trajectory: objective biomarkers emerging for detection, mechanistic bridges from neurons to clinical phenotype, and validated AI systems entering practice with governance frameworks in parallel.

Electronic health records power medical research and AI systems, but conventional validation tools miss semantic inconsistencies. A new framework borrows from software testing to automatically detect when data…

One in five US adolescents use AI chatbots for mental health advice, but most keep it secret. A survey reveals how embedded these tools are in youth mental…

MCG2Vec, a self-supervised learning model, detects multiple cardiac pathologies from magnetocardiography with interpretable patterns aligned to cardiac physiology.

Children who visit the ED for asthma often return. A new machine learning model can identify which ones will, enabling earlier intervention with specialized care.