In the Science study, 304 complex cases published in the New England Journal of Medicine were evaluated without physical examinations or medical imaging available to the AI or human doctors.

Overview Debates Related claims Topics Sources

Running on OpenAI's GPT-5.5 model, MAI-DxO achieved about 85% accuracy on emergency department cases both with and without orchestration.

Microsoft's MAI-DxO system achieved an 80.3% accuracy rate on emergency department admissions, compared to 46.8% for generalist physicians with internet access.

Microsoft's MAI-DxO diagnostic system achieved an 83.2% accuracy rate on complex medical cases from the New England Journal of Medicine in a study published in Science on October 22, 2026.

Microsoft compiled scenarios from medical case reports and emergency department data to create its SDBench benchmark.

Researchers measured how effectively MAI-DxO selected diagnostic tests rather than evaluating only final diagnostic accuracy.

Barrister Peter Mant KC stated at the Lampard Inquiry that clinical effectiveness claims made by Oxevision manufacturer Lio Health were based on unverified anecdotes and unbacked by statistical evidence.

Swedish researchers tested the Dermalyser AI smartphone app across 36 primary care centers from May to December 2022, evaluating 253 suspicious skin lesions in 228 patients.

Researchers trained and validated the FAHR-Face survival prediction tools using photographs of over 30,000 cancer patients with known medical outcomes.

In the Swedish primary care trial, the Dermalyser app achieved an overall area under the ROC curve of 0.960 with 95.2% sensitivity and 84.5% specificity for distinguishing melanoma.

In the U.K. hospital study, the AI device demonstrated sensitivity for malignancy exceeding 97% across all skin cancer types and achieved a specificity of 72.1%.

The FAHR-FaceAge tool demonstrated biological age estimation correlation scores across racial groups of 0.86 for White, 0.81 for Black, 0.86 for Asian, and 0.91 for Hispanic individuals.

FAHR-FaceAge and FAHR-FaceSurvival yield weakly correlated predictions, and combining both measures improved survival-model fit in external validation testing.

Researchers validated the FAHR-FaceSurvival prediction model in an external cohort of nearly 4,900 cancer patients in the Netherlands.

In a validation trial of 3,820 cancer patients, 64 of the 122 patients assigned the highest FAHR-FaceSurvival risk scores died within 90 days, compared to 20 of 975 in the lowest-risk group.

University of Surrey machine learning expert Prof. Miroslaw Bober cautioned that the technology remains a proof of concept that requires further testing across hospitals before clinical deployment.

When evaluated against the AI model TotalSegmentator, the Percival model flagged conditions including pancreatic cancer, cirrhosis, and bacterial pneumonia significantly more reliably.

The findings and technical specifications regarding the Percival model were published by lead author Cameron A. Beeche and senior author Walter Witschey in the journal npj Digital Medicine.

The Tanzanian diabetic retinopathy trial results were published in The Lancet Global Health by lead author Dr. Charles R. Cleland and research colleagues.

A study by the London School of Hygiene & Tropical Medicine evaluated 2,378 people across 24 hospitals in Tanzania to compare AI-supported diabetic retinopathy screening against standard pathways.

The U.S. Department of Energy plans to invest more than $500 million over five years in biological measurement, modeling, and computing for the Virtual Biology Initiative