Detection of Seizures Through Pre-Ictal Phase Using Optimized 1D-Capsule Network

A neurological illness that causes unexpected seizures, epilepsy affects millions of people worldwide. Seizure prediction and therapy greatly depend on the early identification of preictal (before seizure) and interictal (between seizures) EEG signals. In this paper, a unique deep learning architecture called 1D-Capsule Networks for an automated recognition approach has been presented. Capsule Networks improve feature extraction and classification by maintaining spatial links in EEG data, in contrast to conventional CNN models. The Greylag Goose optimizer, which is known for its effectiveness in managing complex neural network topologies, has been employed to optimize the 1D-Capsule Network model. This method is validated using a Children's Hospital Boston dataset, and our results show that it can effectively separate interictal and preictal EEG patterns. The suggested approach attains a high accuracy of 98% and offers perceptions into the temporal dynamics of seizure onset. This discovery shows potential for enhancing the treatment of epilepsy through its innovation in seizure prediction systems.