Detection of Epilepsy Disorder Using Spectrogram Images Generated From Brain EEG Signals

Abstract

Epilepsy (EP) is a persistent neurological condition of chronic brain disorder characterized by repeated seizures and causes psychological issues such as anxiety and depression. There is a need to detect the presence of epilepsy at an earlier stage with the help of technological intervention. Early detection of epilepsy can help medical practitioners treat patients effectively and in a better way. Electroencephalography (EEG) signals are more suitable for monitoring brain activity and detecting brain disorders. In this paper, we propose a deep learning based approach for the early detection of epilepsy via EEG Spectrogram images. The proposed approach is 3-fold. First, we propose an algorithm to generate spectrogram images from the EEG signals, and then, we adapt an efficient Convolutional Neural Network (CNN) model to classify the spectrogram images. Finally, we utilized SmoothGradCAM++ and saliency maps to interpret the decision-making process of the deep learning models. We examined the use of three different pretrained CNN architectures, namely, EfficientNetB0, MobileNetV2, and ResNet18. The methodology is tested on two publicly available datasets to validate the performance of the classifiers in terms of sensitivity, accuracy, specificity, precision, and F1-Score. We observed that the modified MobileNetV2 architecture achieved a state-of-the-art accuracy of 99.24%. The proposed approach can be instrumental in the early detection of epilepsy and other neurological disorders using EEG.

Keywords

• Epilepsy (EP)
• electroencephalography (EEG)
• spectrogram images
• convolution neural network (CNN)
• deep learning (DL)
• MobileNetV2