Designing an Efficient Depression Classification Framework in Healthcare using Residual Recurrent Neural Network with Multimodal EEG Signals

Depression is a mental health disorder and it is one of the leading causes of disability in the world. It often occurs with serious physical conditions such as coronary diseases, stroke, and cancer, depression leads to progress these diseases. However, early detection is a challenging task and most of people do not give importance to depressive symptoms. Moreover, these symptoms can vary from one person to other, so developing effective and comprehensive detection methods is important. One of the neuroimaging tools utilized for diagnosing depression is known as the Electroencephalogram (EEG) and it is useful to identify biomarkers linked to the disorder. Even though the use of EEG signals is beneficial, improvements are still needed to increase the accuracy of diagnosis and clinical usefulness. Researchers have also explored multimodal fusion systems that combine audio, video and text-based features using deep learning techniques to enhance depression detection. However, most of the multimodal approaches have issues like computational complexity. Hence, to tackle all these issues in the classical depression classification framework, a novel deep learning-based depression classification framework is developed. Initially, multimodal EEG signals are collected from benchmark resources. Consequently, the spectral and wave features of EEG signals are extracted for all modes. Further, average feature fusion procedures are performed in the extracted features and then these features are subjected to the developed depression classification phase. Here, Residual Recurrent Neural Network (Res-RNN) is employed to obtain the depression classified outcomes. Later, various analyses are executed in the developed depression classification technique over the classical techniques.