AI Helps Diagnose Rare Diseases That Baffled Doctors

OpenAI’s o3 model contributed to the diagnosis of 18 children with rare diseases at Boston Children’s Hospital, according to a study published Thursday in NEJM AI. The AI analyzed the genomic data of 376 patients, identifying new diagnoses in approximately 5% of cases where previous clinical reviews had failed to find answers.

How did the AI identify these rare diseases?

Researchers analyzed the full DNA sequences, or genomes, of patients and compared them against newly discovered genes. Catherine Brownstein, a lead researcher at the Manton Rare Orphan Disease Research Center at Boston Children’s Hospital, stated that AI could become a significant tool in these processes.

The study included 10 cases of rare neurodevelopmental diseases, four neuromuscular disorders, two cases resulting in sudden death, and two cases of early psychosis. Suyash Shringarpure, a health researcher at OpenAI and author of the study, noted that some cases remained unsolved during initial reviews but were later clarified by subsequent scientific publications.

Did You Know? Rare diseases are estimated to affect approximately 30 million people across the United States.

What was the impact on individual patients?

Kyra Benton is one of the patients who received a diagnosis through AI support. Benton began showing symptoms, including difficulty running and walking on her toes, at age 9, though doctors could not determine the cause for years.

Shortly before she turned 20 last year, researchers diagnosed her with myofibrillar myopathy, a progressive genetic neuromuscular disease. Benton stated she has reservations regarding AI but recognizes its potential benefits in healthcare.

Expert Insight: Samantha Carter suggests that the integration of AI into genomic analysis creates a necessary bridge between massive datasets and clinical application. The value lies not in replacing the physician, but in the AI’s ability to scan updated scientific literature faster than a human could, potentially shortening the “diagnostic odyssey” for families.

Who makes the final medical decision?

The AI did not act as the sole decision-maker in the study. Researchers provided the model with patient symptoms, doctor notes, and genes potentially linked to those symptoms. All results provided by the model were then evaluated and verified by human experts.

What is Openai O3 Rare Disease Diagnosis?

OpenAI’s terms of service explicitly state that its technology should not be used by individuals for self-diagnosis. The final diagnoses in this study were only confirmed after professional human review.

What happens next for AI diagnostics?

This technology may be integrated into more rare disease research centers to speed up the analysis of complex DNA sequences. It is possible that AI could help identify connections between rare symptoms and newly discovered genes more efficiently than manual review.

Future applications may focus on expanding the percentage of successful diagnoses beyond the 5% seen in this study, provided that human experts continue to verify every result.

Frequently Asked Questions

Which AI model was used in the research? The study utilized the o3 model developed by OpenAI, which was released in April 2025.
How many patients were involved in the genomic analysis? Researchers analyzed the genome data of 376 patients with undiagnosed rare diseases.
Did the AI independently diagnose the children? No. The AI suggested findings based on clinical notes and genomic data, but final diagnoses were verified by human experts.

Do you believe AI will eventually become a standard tool for diagnosing rare genetic conditions?