Epileptic Seizure Detection Using Novel CNN with Residual Connections

The paper presents a novel deep neural network-based model for accurate epileptic seizure detection by addressing the challenges of temporal and spatial locality in Convolutional neural networks (CNN) and Recurrent neural networks (RNN)-based architectures. The proposed model incorporates residual networks in a CNN architecture to enhance the feature learning of seizure patterns in the deep neural architecture. The residual connections facilitate the extraction of meaningful information from the raw electroencephalogram (EEG) data without requiring recurrent connections. The utilization of residual networks improves the model's ability to capture both temporal and spatial dependencies, leading to superior performance in seizure detection tasks. The results obtained on the Bonn University dataset demonstrate the model's remarkable performance. In binary classification, the model achieves an impressive maximum accuracy of 99.9% on different testing sets, showcasing the model's effectiveness in capturing spatial and temporal patterns in EEG data. Moreover, the average accuracies obtained for binary and three-class classification are 96.37% and 96.50%, respectively, reaffirming the robustness and generalizability of the model. Additionally, the model exhibits competitive performance in four-class and five-class classifications, achieving average accuracy of 85.75% and 88.25%, respectively. These findings highlight the potential of incorporating residual networks into machine learning-based seizure detection systems and contribute to the existing knowledge in the field of seizure detection. The proposed model's ability to perform remarkably well without the need for preprocessing or feature extraction further underscores its significance in advancing the accuracy and effectiveness of epileptic seizure detection.