More than one in five people who suffer concussions continue experiencing debilitating symptoms months after injury, but new research suggests artificial intelligence could provide advance warning before headaches and fatigue strike. A proof-of-concept study from Norway demonstrates that AI foundation models can predict moderate to severe post-concussion symptoms a full day before they occur by analyzing physiological signals from wearable devices.

Foundation Models Outperform Traditional Machine Learning

Researchers at NTNU and the Norwegian Center for Headache Research tracked 20 patients with persistent post-concussion symptoms over 28 days, collecting daily measurements of heart rate, skin temperature, and trapezius muscle tension. The study, published in The Journal of Head Trauma Rehabilitation, found that foundation models achieved moderate to fairly high accuracy in forecasting symptom onset 24 hours in advance—outperforming traditional machine learning approaches.

Lead researcher Antonios Danelakis from NTNU's Department of Computer Science explained that the findings challenge the perception that post-concussion symptoms occur randomly. "Symptoms follow detectable patterns rather than occurring randomly," the research revealed, suggesting that physiological changes precede conscious symptom awareness.

Study participants were recruited from Rehabiliteringsklinikken at St. Olavs Hospital and logged their data throughout the monitoring period. The research team included collaborators Gøril Storvig, Marte Helene Bjørk, Daniela Contreras, Manjit Matharu, Parashkev Nachev, Bendik Slagsvold Winsvold, Erling Tronvik, Anker Stubberud, and Alexander Olsen.

Addressing a Widespread Clinical Challenge

The research addresses a significant healthcare burden in Norway and beyond. More than 17,000 people visit Norwegian emergency clinics and hospitals annually for minor head injuries, with the true figure likely twice as high when including general practitioner visits and unreported cases. Globally, more than 20% of concussion patients experience symptoms persisting beyond three months post-injury.

Current post-concussion care remains largely reactive, with patients managing symptoms as they arise. The 24-hour prediction window demonstrated in this study could fundamentally shift treatment approaches, enabling proactive symptom management through adjusted treatment plans, anticipatory guidance, and optimized medication timing.

For patients managing chronic post-concussion symptoms, advance warning could enable better daily planning, activity modification, and strategic rest periods before symptom exacerbation. The technology also presents opportunities for wearable device manufacturers and digital health platforms to integrate foundation model-based prediction into consumer products.

From Proof of Concept to Clinical Application

The research team has classified this work as a methodological proof of concept, with plans to expand significantly. A new study launching in spring 2027 will target more than 200 patients using an updated app version, aiming to validate the findings at scale and refine the predictive algorithms.

The study's success with foundation models—AI systems trained on vast datasets that can be adapted to specific tasks—suggests these advanced architectures may be particularly suited to detecting subtle physiological patterns that precede symptom onset, patterns potentially too complex for traditional analytical approaches to capture reliably.

Sources

  1. Foundation Model Forecasting of Headache Days in People With Persisting Postconcussion Symptoms. — The Journal of head trauma rehabilitation · Health AI Central record
  2. Physiological signals help AI predict post-concussion headaches and fatigue — News-Medical.Net