SeizureSight: 2D CNN-LSTM hybrid for EEG-based seizure prediction

This paper introduces SeizureSight, a novel 2D Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) hybrid model designed for EEG-based seizure prediction. Leveraging the spatial feature extraction capabilities of CNNs and the temporal modeling abilities of LSTMs, SeizureSight aims to improve the accuracy and robustness of seizure prediction. The methodology encompasses data selection from the CHB-MIT Scalp EEG Database, rigorous data preprocessing techniques, detailed model architecture design, and comprehensive training and evaluation procedures. SeizureSight’s architecture consists of three CNN layers for spatial feature extraction and two LSTM layers for capturing temporal dependencies. Experimental results on the CHB-MIT Scalp EEG Database demonstrate SeizureSight’s efficacy, achieving an accuracy of 94% in training. Furthermore, qualitative analysis provides valuable insights into the model’s performance and behavior, offering potential clinical implications for epilepsy management.