AI Takes a Role in Diagnosing Rare Illnesses in Children

Artificial intelligence is now a part of our daily lives, and it may be playing a crucial role in saving lives. In an innovative application of this technology, AI has assisted in diagnosing 18 children at a Boston hospital, whose perplexing conditions had left doctors puzzled.

A recent study, published in the New England Journal of Medicine’s NEJM AI, highlights how OpenAI’s o3 model can address complex medical mysteries. Released in April 2025, this AI model has proven effective in catching new diagnoses for patients—including 10 with rare neurodevelopmental diseases, four with neuromuscular disorders, two who had suffered sudden deaths, and two with early psychosis.

Catherine Brownstein, a lead researcher at the Manton Center for Orphan Disease Research within Boston Children’s Hospital, described the findings as “a total game changer,” according to an article from NBC News.

The Manton Center aims to uncover the causes of rare diseases, which affect about 30 million individuals in the U.S. Brownstein indicated that the hospital screens patients’ genomes—essentially their entire DNA set—searching for newly identified genes with potential links to specific diagnoses.

Diagnosing rare diseases is often a lengthy process. Suyash Shringarpure, another study author, noted, “A researcher can only allocate so much time per case.” He pointed out that sometimes, cases that initially remain unresolved later receive clarity due to new research linking genes and diseases.

The researchers examined 376 genomes from patients with undiagnosed rare diseases. Remarkably, the AI identified nearly five percent of new conditions. “Given that these genomes had already been discussed extensively, that’s significant, with each case representing new hope for families,” Brownstein remarked.

One patient, Kyra Benton, benefited from this AI-assisted diagnosis. She began experiencing troubling symptoms at just nine years old, including walking on her tiptoes and struggling with a standard running gait. Over the years, her health deteriorated as doctors attempted to pinpoint the problem. Finally, just before turning 20, she was diagnosed with myofibrillar myopathy, a genetic neuromuscular disorder.

Benton admitted to having initially been skeptical of AI, saying, “I’m not really in favor of AI. But I see that it can have its benefits.”

It’s important to note that OpenAI stresses its technology shouldn’t be used for self-diagnosis. In this study, researchers used the AI model as a supplementary tool, inputting data like doctor’s notes, symptoms, and potential genetic factors. Ultimately, human experts were responsible for reviewing the AI’s suggestions to make a final diagnosis.