Detection and classification of epilepsy using hybrid convolutional neural network

Abstract

In recent years, more than 50 million people have been affected by the epilepsy, neurological disorder diseases. To monitor the situation of the epilepsy patient requires experienced and skilled person. In order to overcome these issues, autonomous detection of electroencephalogram (EEG) signal by deep learning model has evolved. Convolutional neural network (CNN) is one of the sub-category of neural network and widely used in the various field such as weather forecasting, signal processing and medical applications. In this article, the University of California Irvine (UCI) respiratory EEG signals are used to analyse the proposed hybrid CNN and results are compared to the pre-trained GoogleNet Network. EEG signals are initially converted into three different forms such as scalogram, spectrogram and time domain images and classification of images are carried out by the pre-trained GoogleNet network results in an accuracy of 85%. Then time domain images are combined with spectrogram and scalogram EEG signal separately and detection has been carried out by the CNN. It is found that the CNN network yields an accuracy of 92% which was higher than the pre-trained GoogleNet. To enhance the classification accuracy further, scalogram, spectrogram and time domain images are combined as single input images and applied to the CNN network and it results with the accuracy of 98%. The performance metrics such as Sensitivity, Specificity, F1 Score, Precision and misclassification rate of GoogleNet and proposed hybrid CNN networks are evaluated. It is observed from the result that proposed CNN results less than 10% misclassification rate, whereas for GoogleNet it was more than 20%. Similarly, the precision value of GoogleNet and proposed CNN networks are 82% and 93%, respectively.