AI-enabled diagnostic systems outperform human clinicians in medical imaging and patient care across randomized trials.

In the largest randomized trial of medical AI, radiologists using AI support detected approximately 30 percent more instances of breast cancer and 25 percent more invasive tumors than radiologists without AI in a sample of more than one hundred thousand women undergoing mammography. Across forty randomized trials in gastroenterology, machine vision–supported colonoscopy increased the overall detection of adenomas and polyps by more than 20 percent.

An AI foundation model called EchoPrime, trained on over twelve million echocardiographic videos and paired text reports, achieved state-of-the-art performance in interpretation of cardiac features and diagnoses. AI systems can now extract additional disease signals from medical images: chest X-rays can detect type 2 diabetes risk and atherosclerosis risk in the next ten years; retinal images can indicate risk of Alzheimer's, Parkinson's disease, and stroke; and 12-lead electrocardiograms can reveal anemia, diabetes, kidney disease, and atrial fibrillation risk.

A randomized trial in real-world medical centers from four continents confirmed advantages of AI for ophthalmology care in diagnosis of seven major eye diseases and patient management. In a separate randomized trial of primary care at fifteen clinics in Nairobi involving nearly forty thousand patient visits, clinician use of an AI consult tool reduced diagnosis and treatment errors.