AI Model Diagnoses Rare Genetic Diseases With 94% Accuracy, Outperforms Specialists
Amit Yadav
Researchers at Stanford Medicine have unveiled an AI diagnostic model that identifies rare genetic disorders from patient symptom profiles and genomic data with 94% accuracy — significantly outperforming specialist physicians who averaged 72% on the same test cases.
A team of researchers at Stanford Medicine has published results in Nature Medicine demonstrating that their AI model, named RareDx, can diagnose rare genetic diseases with 94% accuracy when given a combination of patient symptom histories and whole-exome sequencing data. The model was tested on 3,200 anonymised patient cases drawn from rare disease registries across the United States and Europe.
Rare diseases — defined as conditions affecting fewer than 1 in 2,000 people — number over 7,000 known variants, and the average patient spends 4.8 years receiving a diagnosis. Many face misdiagnosis multiple times before a specialist correctly identifies their condition. RareDx was specifically trained on phenotype-genotype correlations from the Human Phenotype Ontology database and clinical records, giving it broad coverage across 4,200 rare conditions.
In the controlled study, RareDx was compared directly against panels of specialist physicians including medical geneticists and rare disease clinicians. The AI achieved 94.2% diagnostic accuracy versus the physician average of 72.3% — a gap the researchers attribute to the AI's ability to hold thousands of rare phenotype patterns in memory simultaneously, a feat impossible for any individual clinician.
"This isn't about replacing doctors," said Dr. Euan Ashley, senior author of the study. "It's about giving every patient access to the diagnostic knowledge of a world-class specialist, regardless of where they live or what their insurance covers." The team is working with MeitY in India and NHS Digital in the UK to pilot the tool in under-resourced clinical settings.
The Indian healthcare implications are significant: India has over 70 million rare disease patients, and specialist diagnosticians are concentrated almost entirely in major metro hospitals. A cloud-based deployment of RareDx is planned for AIIMS hospitals by Q3 2026, which could dramatically reduce diagnostic delays for patients in smaller cities and rural areas.