Epileptic seizure detection using posterior probability-based convolutional neural network classifier

Epilepsy is the most common neurological disorders affecting 70 million people worldwide. Nowadays, the advanced Epileptic Seizure (ES) detection from EEG (Electroencephalogram) signal plays a crucial role specifically in its diagnosis process. Traditionally, this diagnosis has been performed by experts using EEG signals on the visual inspection of data for detecting ES. Such a costly and slow process prone to have certain human errors. So, various classification approaches have been developed for epilepsy detection. However, those methods have limitations such as minimum accuracy in detection, high complexity to handle dataset, and low classification rate of features. To handle these challenges, automatic classification system is proposed that classifies the EEG signal as normal, interictal or ictal ones. The features extracted from EEG dataset comprises of time-domain features, frequency domain features and HHTEFs (Hilbert Huang Transform based on Entropy Features). Then, the hybrid correlation filter method combining the KRPC (Kendall Rank and Pearson Correlation) method is employed to select only the significant features and essential in extraction process for improving the classification performance, thus by obtaining relevant features to ease feature classification. The selected features undergo the classification task by using proposed PP-CNN (Posterior Probability-based Convolutional Neural Network). Experimental analysis proved the performances of the proposed method in terms of accuracy, sensitivity, specificity, precision, recall and F1-score. From performance analysis, the proposed framework yielded effective detection for non-seizure and seizure patients than existing methods acquiring 99% accuracy, 99% sensitivity, 99% specificity, 98.50% precision, 99% recall and 98% as F1-score rate.