Personalized Seizure Detection Using Spiking Neural Networks

Around 50 million people worldwide suffer from Epilepsy, making it one of the most common neurological diseases. Epilepsy is characterized by sudden intermittent seizures, often imposing profound social and physical limitations. More importantly, the risk of premature death in these patients is up to three times that of the corresponding healthy population. It is estimated that 30% of the population with epilepsy is at risk of severe trauma, or premature death, despite currently available treatments. Smart wearable systems could mitigate many of the risks associated with epilepsy and seizures by providing early warnings for the patients and caretakers to take precautions. However, wearable systems are highly constrained in terms of resources and, therefore, are generally unable to utilize modern machine learning, due to their limited computing power, memory storage, and energy/battery budget. To address this issue, in this paper, we consider personalized seizure detection by adopting spiking neural networks, which are known to be efficient in terms of energy. Our experimental results demonstrate that our personalized spiking neural networks are on par with their artificial neural network counterparts in terms of performance, reaching a sensitivity of 78.8 % and a specificity of 76.9 %.