A prospective real-world study in nuclear medicine found that ChatGPT v4.1 performed well on administrative patient queries, but its medical answers still carried clinically relevant risks.

Single-night polysomnography can misclassify obstructive sleep apnea when sleep varies substantially from night to night. This study suggests that zero-burden, multi-night home monitoring may help reveal that variability.

A multicenter study evaluated AIOC, an AI-assisted ultrasound system for fetal orofacial cleft diagnosis, showing senior-level performance and measurable gains for junior radiologists.

Large language models may help integrate passive smartphone data to detect changes in depressive state, but current performance remains modest.

A multicenter MRI study suggests that knee bone shape is not just background anatomy: it may be a significant intrinsic marker of ACL injury susceptibility.

A zero-shot retrieval-augmented LLM pipeline generated structured thoracic oncologic histories with high factual performance and much faster turnaround than manual chart review.

A new preoperative AI model estimates cancer-specific mortality in surgically treated, nonmetastatic renal cell carcinoma using eight routinely available clinical variables.

Metabolic bariatric surgery remained uncommon among commercially insured US adults with newly diagnosed obesity and BMI of at least 35, despite established clinical eligibility. In this cohort, the…

A multicenter Radiology study shows that deep learning can detect both pancreatic masses and subtler indirect CT signs linked to pancreatic cancer.

Recent AI & Computer Science coverage shows medical AI moving toward deployable biomarkers, multimodal oncology reasoning, workflow automation, and LLM stress-testing under realistic clinical constraints.