Spiking Neural Networks and Temporal Transformers for Real-Time Seizure Prediction and Cognitive Brain-Computer Interfaces

Abstract

Prediction of epileptic seizures using Electroencephalography (EEG) is an important research area, as it can provide early warning and enhance patient safety in neurological healthcare. Deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Vision Transformers and Spiking Neural Networks (SNNs) have achieved great success. However, the existing methods are often limited by long-range temporal dependencies, event-driven neural computation, functional brain connectivity modeling and interpretability. To address these limitations, in this paper, we propose a novel hybrid framework NeuFormer-SNN for real-time seizure prediction and cognitive Brain-Computer Interface (BCI) applications. The proposed model is based on Hierarchical Temporal Transformers (HTT) to learn multi-scale temporal representations from EEG signals, which is capable of effectively capturing local neuronal dynamics and long-term dependencies. These representations are then converted into sparse spike trains with a Hybrid Spiking Neural Network (HSNN) that supports biologically inspired and energy efficient computation. We propose an Adaptive Spike Attention Module (ASAM) to selectively attend critical spike events; EEG channels and temporal segments associated with the seizure progression to further improve the performance. In addition, a Dynamic EEG Graph Attention Network (DEGAT) is proposed to model functional brain connectivity with attention-driven graph learning. A Neuro-Symbolic Cognitive Reasoning (NSCR) module is used to combine learned representations with expert neurological rules to produce explainable predictions with confidence estimation for clinical interpretability. The experimental results demonstrate its advantages in accuracy (99.31%) and efficiency, and validate that NeuFormer-SNN is a promising and explainable solution for next-generation seizure prediction system.