All traces

An ensemble of machine learning models analyzed 2.4 million physiological samples from simulator drives to classify driver cognitive states, with XGBoost identified as the most robust classifier.

A Random Forest trained on 22 routine clinical variables predicted unplanned 90-day psychiatric readmission in adults with MDD with moderate discrimination.

In NSCLC immunotherapy selection, CB-only AI models outperformed PD-L1 and clinical scores in the independent test set, and both expert and nonexpert physicians improved their predictions when using the explainable AI decision support tool.

AI systems including large language models can process unstructured clinical text and interpret complex laboratory findings to support clinical decision-making in laboratory medicine.

Large language models have the potential to democratize medical knowledge and facilitate access to healthcare in clinical practice.

Standardizing pre-analytical H&E preparation before AI deployment can reduce computational burden and deployment costs while supporting robust AI performance in pathology.

A ResNet-18-based model trained on neonatal ICU radiographs detected pneumothorax on supine chest radiographs with AUC 0.975, 87.6% sensitivity and 95.3% specificity in the test set.

Systematic review found most ML algorithms improved CVD risk prediction compared with Framingham Risk Score by incorporating sociodemographic predictors and modeling non-linear interactions.

AI systems are being developed to fuse facial expressions, voice, and physiological signals to objectively quantify pain and to automatically segment spine, nerves, and needle tips to improve identification accuracy.

AI-assisted orthodontic systems and remote monitoring platforms reduced in-person appointments while maintaining clinical standards and improving patient compliance in included studies.

ARIA's ruling that chart-eligible recordings must be substantially human-made protects human artists and preserves chart integrity for human creativity.

Roles centered on human interaction, responsibility and physical oversight in medicine, nursing, engineering and education are expected to persist because people prefer humans for care, difficult news and large-scale physical work.

An AI system combining text normalisation with 1931 features maintained stable self-harm detection in prospective validation at its development metropolitan hospital, achieving PR AUC 0.84 over 329,655 triage notes in the following four years.

Explainable ML models achieved near-perfect discrimination for breast cancer diagnosis on cytology data, with top models reaching 0.996 AUC and 98.25% accuracy, supporting use in resource-constrained diagnostic workflows.

Large language models including ChatGPT-o3, ChatGPT-5.2, Gemini 3 and DeepSeek provided generally acceptable clinical accuracy when answering 30 frequently asked patient questions about robotic-assisted total knee arthroplasty.