Scientists develop AI headset to predict epilepsy seizures

Researchers at Glasgow Caledonian University have developed an AI-powered headset that analyses brainwaves and heart signals to warn of seizures, with reported accuracy of up to 95%; the team says wider availability will take several years because of further testing and regulatory approvals.

Researchers at Glasgow Caledonian University in Scotland have developed an AI-powered headset designed to predict epileptic seizures minutes before they begin. The device analyses brainwave (EEG) and heart (ECG) signals to detect patterns that precede seizures. Lead researcher Professor Hadi Larijani said even a few minutes' warning could help patients and families prepare and avoid injuries. Epilepsy affects around 630,000 people in the UK and can include many different seizure types. The team says the headset is still being refined and will need regulatory approvals before it can be widely used. Key details: - The headset uses EEG and ECG signals to identify electrical and physiological changes that can occur before a seizure. - Researchers trained the AI with thousands of hours of historical EEG and ECG recordings and tested it on a custom "phantom head" to simulate real use. - The team reports the system can achieve up to 95% accuracy and that the algorithm can display its confidence level about a predicted seizure. - The researchers aim for a discreet, wireless, light, cap-like wearable and hope future models will be suitable for children. - The project received support from UK Research and Innovation through a £9 million Proof of Concept award. - Further hardware refinement, clinical evaluation and medical-device accreditation are planned, so wide availability is likely to take several years. Summary: Researchers say the headset could give people with epilepsy and their families minutes of advance warning, which may help prevent injuries and reduce fear. The system has shown promising accuracy in early testing, but further development, clinical studies and regulatory review are expected. A date for wide availability is undetermined at this time.