Purpose: Tonic-clonic seizures can be very dangerous, and may substantially increase the risk of SUDEP. One way to potentially mitigate this is to detect these seizures in real-time, ensuring immediate medical attention. However, only a few people on a global scale have access to real-time seizure detection devices. Therefore, there is a need to develop real-time seizure detection devices on affordable hardware, to reach larger populations.
Method: In this work we developed LAMPSY, a seizure detection device designed to be unobtrusive, accurate, affordable, and privacy-preserving. LAMPSY is discreetly built into a light fixture, thereby removing the need for constant user interaction or uncomfortable sensors, and reducing stigma and impact on the user’s daily life. LAMPSY's algorithm, based on Optical Flow, Principal Component Analysis, Independent Component Analysis and Machine Learning classification, was tested on a dataset of 21 tonic-clonic seizures from 12 patients.
Results: LAMPSY's algorithm was able to detect tonic-clonic seizures with a sensitivity and specificity of 99.06% ± 1.65% and an average latency of 37.45 ± 1.31 s. Moreover, LAMPSY was able to run in real-time on a Raspberry Pi 4B, ensuring affordability and compactness.
Conclusion: This work extends the state-of-the art in video-based seizure detection by demonstrating a successful real-time implementation that does not require expensive hardware, and highlights the potential of integrating technology seamlessly into everyday environments and in resource-restricted countries.