Classification of Major Depressive Disorder Based on EEG Signals Using Superlet Transformation and ResNet-18

Major depressive disorder is an evolving mental health issue considered by WHO as one of the main contributors to the global disability. It has been affecting nearly 350 million people around the worldwide along with the highly increasing rate of suicidal attempts. Electroencephalogram (EEG) is a non-invasive method that detect the neural activity in the brain which reflects the working status of the human brain. It is used as the diagnostic tool for the depression detection. In this proposed work the dataset includes the EEG signals of 64 subjects (30 Healthy controls and 34 MDD patients). Each EEG signals have collected the signals from the 19 electrode channels. This project focus on the classification of the EEG signals of MDD individual from the Healthy using the deep learning algorithms. The deep learning algorithms effectively helps to find the patterns between the healthy and MDD individuals. The proposed model implements the Independent Component Analysis and Wavelet Denoising for cleaning the noisy signals and removing the irregularity from the raw EEG signals. The cleaned EEG data is transformed using the Superlet Transformation, the transformed data is converted into the images where all the 19 channels signal is compressed into a single image by taking the mean of all the channels. The transformed images are given to the pretrained deep learning model-various ResNet models for the feature extraction and classification. The ResNet-18 architecture achieves the highest accuracy of 95.65%. Which effectively have the impact on the early diagnosis of the MDD patients.