Epileptic Seizures Detection from EEG Signals using Improved Hybrid RNN-CNN Feature Extraction Approach

Abstract

Epileptic Seizures(ES), a neurological condition, presents significant challenges in both diagnosis and management. This is crucial in ensuring early detection and treatment before the negative impacts surface and improving patient experiences. In this paper, an automatic method implemented using deep learning hybrid architecture (CNN-RNN) integrates “feature extraction” methods such as DWT, LBP, EMD, and FIT to detect ES from EEG signals. The suggested approach deals with the EEG signals. Thus, CNNs are used to extract the spatial features, and LSTMRNN considers the temporal features of EEG data and model, therefore, handles the short-term and long-term features of the EEG data effectively.

Evaluation of the planned hybrid CNN-RNN model shows promising results surpassing those of MultiSVM by a considerable size in the accuracy measure (98.38%), precision (83.24%), and recall (83.25%) as well as specificity of 95.81% and F1 score of 82.61%. By only having 4.19% false positives and an average specificity, the decisions made by this model are poised to accurately identify clinical seizures in real-time with the minimized likelihood of wrong classification.

This method could be used for continuous, binary seizure identification and integrated into wearable EEG monitoring for constant use. Further work should be done on improving the model, including some effects of noisy data and inadequacy of data sets, and extending the model across a patient's different sexes for better applicability of such a system.