Healthcare AI Daily Briefing

Medical Imaging

AI biophysical model predicts HCC ablation volume more accurately than manufacturer estimates.

Journal article (European radiology experimental)

In a retrospective study of 96 RFA procedures across 65 HCC patients treated with a single cluster-needle ablation, a fully automated GPU-accelerated biophysical model integrating patient-specific anatomy and vasculature achieved a median relative volume difference of 0.39 (IQR 0.24–0.57) versus 1.04 (0.64–1.61) for manufacturer estimates (p<0.001). Simulated volumes matched ground truth across 94 evaluable cases (p=0.165), while manufacturer algorithms systematically overestimated ablation volume; incorporating patient-specific ablation time further improved accuracy. [1]

Policy & Governance

STAT+ reports FDA and CMS plan state collaboration on health AI.

STAT+ reports that the FDA and CMS are looking to collaborate with states on health AI policy, according to its Health Tech newsletter. The article is exclusive to STAT+ subscribers; the newsletter covers FDA regulation of AI, Medicare payment for health technology, clinical AI use, mental health chatbots, and consumer wearables. [2]

Research & Evidence

AI in dialysis reaches implementation milestones with first RCT and society framework.

Journal article (Current opinion in nephrology and hypertension)

A randomized trial showed AI-guided erythropoiesis-stimulating agent dosing achieved outcomes comparable to physician-directed dosing in maintenance hemodialysis, while a meta-analysis of 21 randomized trials supports technology-guided dry-weight adjustment. The first professional society framework for responsible AI in kidney care and FDA lifecycle guidance for AI-enabled devices mark a shift from algorithm development to clinical implementation, with workflow integration and diverse-population validation now the primary challenges. [3]

Clinical Practice

Removing race from clinical tools shows measurable outcome improvements across specialties.

KFF analysis documents that eliminating race-based corrections from eGFR equations led to earlier CKD diagnosis and more kidney transplants for Black patients, while revised lung function equations reclassified more Black patients with severe lung disease. The PREVENT cardiovascular risk calculator performs as well or better without race, and a 2024 survey found 50% of medical societies are working to eliminate harmful race-based algorithms, though federal funding cuts threaten continued impact research. [4]

Policy & Governance

Israeli medical center AI triages mental health patients for urgent care.

Podcast from KFF's The Business of Health with Chip Kahn

The interview describes an AI avatar called Live, developed at an Israeli medical center and now commercialized as Mentally, that converses with patients seeking mental health appointments to prioritize urgent cases for immediate specialist access. The system is regulated in Israel and undergoing testing with the U.S. Department of Defense and VA for FDA submission, adding triage capacity without replacing clinicians. [5]

Industry & Products

Mayo Clinic Platform launches AI chatbot for clinical decision support across 35 specialties.

Clinicians can now query an AI chatbot built on Mayo Clinic's medical knowledge base for diagnosis and care decisions across more than 35 specialties, with integration into several electronic health records for personalized care plans. The tool includes 22 clinical calculators across cardiology, oncology, hepatology, and endocrinology, and was developed with clinician testing and feedback as part of Mayo's strategic AI partnerships including Microsoft. [6]

Research & Evidence

Machine learning identifies 10.5% preoperative weight loss threshold predicting bariatric outcomes.

Journal article (Obesity surgery)

In a cohort of 802 adults undergoing metabolic and bariatric surgery, machine learning models with nested cross-validation found preoperative total body weight loss consistently among the strongest prognostic features for multi-year outcomes. Linear mixed-effects modeling confirmed a 10.5% PreopTBWL% threshold as an independent predictor of durable postoperative weight loss after adjustment for diabetes and operative duration, with procedure type not modifying this association. [7]

Medical Imaging

Two-level AI model detects ischemic pneumatosis more accurately than radiologists.

Journal article (APL bioengineering)

In a retrospective study of 612 patients from three tertiary hospitals, a two-level AI system combining EfficientNet-B3 x-ray screening with 3D nnU-Net CT segmentation achieved an internal test AUC of 0.947 for ischemic pneumatosis, outperforming senior radiologists' 0.876, with external validation AUC of 0.931. SHAP analysis identified portal venous gas and blood lactate as the top predictors. [8]