For decades, diagnosing migraine has relied on a doctor's ear and a patient's description of pain. There is no blood test, no scan, no definitive biomarker. Clinicians piece together symptoms—throbbing head pain, nausea, sensitivity to light and sound—and make a judgment call. But new research from Norway suggests that migraine leaves a much wider biological footprint, one that artificial intelligence can detect even when headache symptoms are deliberately withheld.
Scientists at the Norwegian University of Science and Technology (NTNU) in Trondheim used machine learning to identify people with migraine from a broad set of health, lifestyle, and genetic data—without ever being told about their headaches. The model, trained on information from more than 43,000 participants in the Trøndelag Health Study, distinguished migraine sufferers from controls with an accuracy score of 0.80, a strong result for a condition with no established biological test.
“The fact that AI could identify migraine so accurately without knowing anything about the headache itself suggests that the condition leaves traces that extend far beyond the attacks,” said Anker Stubberud, a physician and headache researcher at NTNU and co-author of the study published in the journal Neurology.
Beyond the headache
The model drew on dozens of variables: demographics, mental health, cardiovascular and musculoskeletal conditions, sleep patterns, exercise habits, medication use, and genetic markers. Age emerged as the strongest predictor, followed by neck pain, menstruation, and nausea. Adding genetic data improved the model only marginally over clinical information alone, suggesting that migraine's systemic effects are already encoded in everyday health records.
“There are many inconsistencies with the diagnosis of migraines. It appears that AI could help us make the correct diagnosis,” Stubberud said.
The study then took a second, more ambitious step. Instead of asking the algorithm to separate migraine from non-migraine headache, the researchers let it find natural groupings within the data. Among 12,185 people with complete headache information, the algorithm identified a cluster of 1,425 individuals, 94% of whom met migraine criteria. That migraine-like group split into four distinct subtypes.
Four faces of migraine
One subgroup consisted exclusively of men. Another was defined by prominent neck pain. A third showed higher levels of musculoskeletal pain alongside anxiety and depression. The fourth resembled what clinicians would call “classic” migraine, with patients experiencing aura—temporary neurological symptoms like flashing lights, zigzag patterns, or tingling that precede or accompany an attack.
These subgroups also differed genetically. Machine-learning-based genetic risk scores distinguished the groups more effectively than conventional polygenic risk scores, reinforcing the idea that migraine is not a single disease but a family of related conditions.
“This strengthens the hypothesis that migraine is not a single disease, but rather a diverse group of different biological conditions,” Stubberud said.
The implications extend beyond diagnosis. If different migraine subtypes have distinct biological signatures, doctors could one day predict which patients will respond to which therapies, moving away from the current trial-and-error approach. “So that people can receive the best possible treatment,” Stubberud added.
For the estimated one in seven Europeans who live with migraine, that would be a significant shift. The condition is a leading cause of disability worldwide, yet it remains underdiagnosed and undertreated. As AI continues to reshape medicine—from genetic risk for cancer to eating disorders—this study offers a glimpse of a future where neurological conditions are defined by biology, not just symptoms.
The Norwegian team's work is a reminder that the most common disorders often hide the most complexity. Migraine may be more than a headache—it may be a spectrum of conditions waiting to be untangled.
