Lampsy: An Invisible Real-Time Epilepsy Video Monitoring and Automatic Seizure Detection Device

Epilepsy causes potentially fatal seizures, leading to fear and anxiety among people with epilepsy and caregivers. Real-time seizure detection can notify caregivers of seizures, helping reduce anxiety. Most available seizure detection devices use visible sensors, possibly preventing widespread adoption due to the stigma still associated with epilepsy. “Invisibles,” i.e., off-the-person devices seamlessly integrated into daily life, may offer an unobtrusive alternative, with video monitoring being a promising modality. However, no video-based, medically certified seizure detection devices exist, possibly due to the hardware costs needed for accurate real-time video-based detection. This work seeks to solve this problem by exploring the feasibility of real-time video-based seizure detection on affordable edge devices. For this purpose, we developed Lampsy, a privacy-preserving video-based seizure detection device embedded within a light fixture. Lampsy’s previously published detection algorithm, which employed Optical Flow, achieved a sensitivity and specificity of 99.06% $\pm ~1.61$ % for 21 tonic-clonic seizures, but required calibration and significant processing, rendering it impractical for real-time use on edge devices. Using the same dataset, we tested various Optical Flow methods and optimizations with the goal of achieving real-time detection on a Raspberry Pi. We achieved a real-time performance of 30 FPS and a sensitivity and specificity of 99.76% $\pm ~0.35$ % without calibration. Lampsy achieves accurate real-time video-based seizure detection on Raspberry Pi edge devices. This work extends the state-of-the art by demonstrating real-time video-based seizure detection on affordable hardware, highlighting the potential of integrating advanced digital health technology seamlessly into everyday environments.