Patients turn to AI to solve rare-disease mysteries
The Wall Street Journal this weekend documents how AI has quietly become a fixture of the rare-disease diagnostic hunt — driven as much by desperate families as by hospitals.
Patients, families, doctors and nurses are increasingly turning to AI tools like Face2Gene to identify rare and hard-to-diagnose diseases, according to a Wall Street Journal report. People with rare conditions often spend five years or more searching for a diagnosis; the Journal's piece walks through cases where AI compressed that search dramatically. Face2Gene, an AI tool that analyzes a patient's facial features and clinical details against databases of genetic syndromes, is the best-known example — and it has enough clinical validation behind it that an international case series in the American Journal of Medical Genetics recently described how the software changed real-world testing strategy in real time.
The human side is what makes the story land. Rachel Hinken spent years watching her son Oliver miss growth milestones — speech and walking delays, height stuck between the zero and first percentiles, barely four feet tall at age ten — while doctors insisted he was fine and would catch up. In another case, an AI reading of a 77-year-old's ECG flagged a 98% chance of rare cardiac amyloidosis, a diagnosis his own doctor had never made. "If it wasn't for AI, I might have been treated for something else," the patient said. "Maybe I wouldn't be here, who knows?" Cardiologist Eric Topol noted that Mayo Clinic now runs AI ECG reads routinely — one of very few health systems to do so.
The notable pattern here is bottom-up adoption: AI is entering rare-disease care through families who bring the tools to their doctors, not just through hospital IT roadmaps. The AI rarely delivers the diagnosis on its own — it flags candidates and suggests which tests to run, letting clinicians confirm. But that is exactly where the leverage sits, because the bottleneck was never clinician effort; it was the search space. Multimodal models that read imaging, genetic markers and phenotypic signals at once turn a brute-force hunt into a prioritized shortlist. The caveats are real: tools like Face2Gene are aids, not oracles, and a confident wrong suggestion in the hands of a desperate family cuts both ways. Still, for families stuck in years-long diagnostic limbo, the Journal's reporting suggests the answer is increasingly arriving faster.
What to watch: whether health systems start folding AI phenotyping and ECG analysis into standard practice the way Mayo Clinic has, and how regulators treat tools that patients — not physicians — bring to the exam room.
Would you let an AI weigh in on a diagnosis your doctors couldn't crack? Tell us in the comments.
