New AI Approach Monitors Brain Health Using Passive Wearable Data

Brain health spans cognitive and emotional functions and can fluctuate even in adults without diagnosed disease. Detecting early changes remains difficult in routine care and burdens specialty services as populations age. Researchers now report that data from smartwatches and smartphones can help monitor these fluctuations using artificial intelligence (AI). The approach aims to support earlier recognition of changes relevant to neurological and psychiatric care.

Researchers at The University of Geneva developed an IA approach that uses passively collected signals from a smartwatch and a dedicated smartphone app to assess day-to-day brain health. The system integrates heart rate, physical activity, sleep patterns, and contextual exposures. It is designed for continuous, noninvasive monitoring without altering daily routines. The goal is to anticipate short-term cognitive and emotional variability.

Eighty-eight volunteers aged 45 to 77 wore the devices for 10 months. The devices captured passive data across 21 indicators, including weather conditions and air pollution levels in addition to physiologic measures. Participants also provided “active” data every three months through standardized questionnaires assessing emotional state and cognitive performance tests. These active measures served as reference points for validating AI predictions.

After data collection, the passive signals were modeled with AI and compared against the questionnaire and test results. The average prediction error across outcomes was 12.5%. Emotional states were predicted more accurately than cognitive states, with error rates generally between 5% and 10% versus 10% to 20%, respectively.

Feature analyses indicated that air pollution, weather conditions, daily heart rate, and sleep variability were the most informative factors for cognitive outcomes. For emotional states, weather, sleep variability, and heart rate during sleep were most influential. These findings suggest that environmental and sleep-related inputs meaningfully contribute to short-horizon brain health forecasts.

The research was conducted across the Geneva School of Economics and Management’s Research Institute for Statistics and Information Science and the Faculty of Psychology and Educational Sciences’ Cognitive Aging Laboratory. It forms part of the joint faculty project Providemus alz. Results were published in npj Digital Medicine.

The next phase will extend monitoring to 24 months and examine characteristics linked to the highest- and lowest-performing AI models. The team indicates this could clarify real-world applicability for individualized monitoring. The approach may open avenues for earlier detection of abnormalities or changes in brain health.

"The aim was to determine whether AI could predict fluctuations in participants' cognitive and emotional health based on these data," said Igor Matias, doctoral assistant at the Research Institute for Statistics and Information Science at the Geneva School of Economics and Management at the University of Geneva and lead author of the study. "On average, the error rate was just 12.5%, opening up new possibilities for the use of connected devices in the early detection of abnormalities or changes in brain health".