IMPROVED FEATURE-LEVEL FUSION APPROACHES WITH MULTIMODAL EEG SIGNAL FOR DEPRESSION CLASSIFICATION USING ADAPTIVE GATED RECURRENT UNIT

Depression is the major cause of increased number of suicides worldwide. Therefore, proper diagnosis and treatment are essential to reduce the impact of depression. An instrument named Electroencephalogram (EEG) is utilized to test and analyze the electrical activity of the brain. Also, it can be used to make an accurate depression severity report. Previous research has demonstrated that it is possible to use deep learning models and EEG information to detect mental disorders. Nevertheless, there are other problems in diagnosing depression, including poor accuracy, subjectivity bias, and limited patient cooperation. Therefore, successful depression classification requires a reliable and unbiased assessment approach. Hence, a multimodal EEG data-based depression classification framework is developed to solve the issues of the classification process using individual modality EEG data for depression diagnosis. In this paper, the input EEG signals are collected from three different electrode positions on the scalp neutral, positive, and negative audio stimuli to boost the depression recognition accuracy. Then the EEG signals with different modalities of deep features are extracted using Autoencoder (CAE) and then these extracted features are fed to Multiscale Feature Fusion-based Adaptive Gated Recurrent Unit (MFF-AGRU) for feature fusion and recognition of depression. The feature fusion process generates powerful features for the final recognition process. By fusing features from multiple modalities, the resulting fused features increase the recognition power of this proposed model. The performance of this suggested framework is enhanced by fine-tuning the parameters of GRU using the Enhanced Random Functions-based Fennec Fox Algorithm (ERF-FFA). Finally, the experimental analysis is done to ensure the data quality, and processing capability of EEG signals for the classification of depression.