A study comparing physician-based surprise question, AI-driven GEST model, and a combined approach shows improved accuracy and efficiency in predicting 6-month mortality among older ED patients.
A machine learning algorithm can generate synthetic CT images from temporal bone MRI, enabling visualization of bone and soft tissue simultaneously without radiation, useful for otologic surgical planning and localization.
In a 14-institution cohort of more than 508,000 operations, MySurgeryRisk maintained high discrimination for ICU admission, mechanical ventilation, acute kidney injury, and in-hospital mortality using routine electronic health record data.
A fine-tuned ensemble of tabular foundation models moderately predicted pathological complete response after TNT and showed potential to support watch-and-wait selection after recalibration.
In a large Veterans Health Administration cluster randomized study, CADe availability increased adenoma detection during colonoscopy without changing withdrawal time or nonadenomatous lesion detection.
Unsupervised machine learning identified three distinct preeclampsia phenotypes linked to timing of delivery, placental dysfunction, fetal growth, and complications, suggesting a path toward more personalized risk assessment and management.
A major 2026 study suggests donor age affects survival differently in haploidentical and matched unrelated stem cell transplants, offering a more practical framework for choosing the best donor.
A qualitative study found that ChatGPT 3.5 described “great surgeons” differently by demographic identity, echoing stereotypes that may shape evaluations, leadership perceptions, and inclusion in surgery.
In Kenyan outpatient clinics, an AI-enabled ECG algorithm showed strong discrimination for left ventricular systolic dysfunction, with 95.6% sensitivity and 99.1% negative predictive value against echocardiography.
In a real-world academic emergency department, ambient AI scribe use was uncommon and concentrated among a small group of physicians, but when used it was associated with shorter documentation and total EHR time.
A multi-institutional AI model identified pre-diagnostic pancreatic cancer on routine CT with meaningful lead time, higher sensitivity than radiologists, longitudinal stability, and externally validated specificity in a low-prevalence setti
A cross-sectional evaluation of AI-generated clinical notes found significant quality deficits compared to human-produced notes, particularly in thoroughness, organization, and usefulness.
Researchers developed and validated ADAPT-CEC, an artificial intelligence algorithm that can adjudicate cardiovascular events across different trial definitions. A hybrid approach combining AI with selective human review achieved 95.6% accu
A multicenter randomized controlled trial reveals that integrating action observation with virtual reality significantly improves paretic hand dexterity in stroke patients, with benefits persisting six months post-intervention, offering a s
This review evaluates the breakthrough ECG2HF model, a publicly available AI tool that outperforms standard clinical scores in predicting 10-year heart failure risk using 12-lead ECG waveforms.
A multicenter study in JAMA Ophthalmology introduces a deep learning model using domain-shift technology to accurately detect macular diseases across different OCT hardware vendors, achieving high negative predictive value and establishing
The MASAI trial demonstrates that AI-supported mammography screening achieves higher sensitivity and non-inferior interval cancer rates compared to standard double reading, while maintaining specificity and significantly reducing radiologis
The TRICORDER trial demonstrates that while AI stethoscopes are capable of detecting heart failure, atrial fibrillation, and valvular disease, their real-world implementation in UK primary care did not yield a statistically significant incr
A multicentre randomized controlled trial demonstrates that robotic telesurgery over distances up to 2800 km is non-inferior to local surgery for urological procedures, maintaining high success rates, low latency, and equivalent safety prof
A crossover study involving 30 ophthalmologists and trainees reveals that while Claude-3.5-Sonnet collaboration improves diagnostic accuracy in complex cases, it also increases confidence in incorrect decisions and fails to reach the perfor
Open language-specific specialty feeds and department pages.