Automatic and Efficient Framework for Identifying Multiple Neurological Disorders From EEG Signals

The burden of neurological disorders is huge on global health and recognized as major causes of death and disability worldwide. There are more than 600 neurological diseases, but there is no unique automatic standard detection system yet to identify multiple neurological disorders using a single framework. Hence, this study aims to develop a common computer-aided diagnosis (CAD) system for automatic detection of multiple neurological disorders from EEG signals. In this study, we introduce a new single framework for automatic identification of four common neurological disorders, namely autism, epilepsy, parkinson's disease, and schizophrenia, from EEG data. The proposed framework is designed based on convolutional neural network (CNN) and spectrogram images of EEG signal for classifying four neurological disorders from healthy subjects (five classes). In the proposed design, firstly, the EEG signals are pre-processed for removing artifacts and noises and then converted into two-dimensional time-frequency-based spectrogram images using short-time Fourier transform. Afterwards, a CNN model is designed to perform five-class classification using those spectrogram images. The proposed method achieves much better performance in both efficiency and accuracy compared to two other popular CNN models: AlexNet and ResNet50. In addition, the performance of the proposed model is also evaluated on binary classification (disease vs. healthy) which also outperforms the state-of-the-art results for tested datasets. The obtained results recommend that our proposed framework will be helpful for developing a CAD system to assist the clinicians and experts in the automatic diagnosis process.