Area-Efficient Multiplier-less Neuromorphic Hardware for Epileptic Seizure Detection Using NSFD-Discretized Spiking Neural Networks

This paper introduces a spiking neural network (SNN) model for epileptic seizure detection, employing non-standard finite difference (NSFD) discretization techniques to overcome stability issues associated with traditional methods like Euler and Runge-Kutta. Applied to a Leaky Integrate-and-Fire (LIF) neuron model, the NSFD approach improves accuracy and stability. The model is implemented on a high-speed, area-efficient, multiplier-less neuromorphic hardware architecture. Trained on EEG data from 400 patients and tested on 100 patients, the system utilizes weight-fused features derived from a Binary Battle Royale algorithm. The weights are stored in block RAM for real-time seizure classification, achieving nearly 99.47% accuracy. The proposed architecture is optimized for real-time, resource-constrained environments, making it ideal for wearable devices in continuous seizure monitoring. This invention represents a breakthrough in neuromorphic computing and medical signal processing, offering a scalable, accurate, and efficient solution for seizure detection in neurological diagnostics.