An efficient CNN based epileptic seizures detection framework using encrypted EEG signals for secure telemedicine applications

Abstract

Recently, the rapid development of Artificial Intelligence (AI) applied in the Medical Internet of Things (MIoT) for the diagnosis of diseases such as epilepsy based on the investigation of electroencephalography (EEG) signals. Thanks to AI-based deep learning models, the procedure of epileptic seizure detection can be performed professionally in Smart Healthcare. However, the security issues for protecting sensitive medical EEG signals from disclosure and unauthorized operations from severe attacks over open networks. Therefore, there is a serious need for providing an effective method for encrypted EEG classification and prediction. In this paper, a new and efficient encrypted EEG data classification and recognition system using Chaotic Baker Map and Arnold Transform algorithms with Convolutional Neural Networks (CNNs). In this system, the channel's EEG time series is first converted into a 2D spectrogram image and then encrypted using Chaotic Baker Map and Arnold Transform algorithms, and finally fed to CNNs-based Transfer Learning (TL) models. From the experimental results, the proposed approach is validated and evaluated on a public CHB-MIT dataset and the googlenet with encrypted EEG images provides satisfactory performance by outperforming the models of other CNN like Alexnet, Resnet50, and squeezenet.