MEMS resonator-based reservoir computing for epileptic seizure prediction
Shuto Kawaguchi, Amit Banerjee, Jun Hirotani, Toshiyuki Tsuchiya · 2024
We developed a microelectromechanical system (MEMS) resonator-based physical reservoir computing (PRC) system for epileptic seizure prediction by real-time classification of electroencephalogram (EEG) signals. We use Si MEMS resonators (width ~ 1 μm, length ~ 100 μm), fabricated by conventional clean-room microfabrication tools, and implement them as low power consuming information processors. The purpose is to realize a portable and affordable epileptic seizure prediction device that demands significantly lower training and computational cost than conventional artificial intelligence-based approaches. In the physical implementation of this method, we successfully classified ‘normal’ and ‘seizure’ type EEG signals with 91.1 % accuracy, and ‘normal’ and ‘pre-seizure’ type signals with 77.1 % accuracy. We expect these results will contribute to the successful development of automatic seizure detection and prediction devices in the future, which will help avoid accidents and injuries in epileptic patients.