PsychNet: Explainable Deep Neural Networks for Psychiatric Disorders and Mental Illness

Mental disorders are irreversible that cause disturbance in behavior, sleep, and emotional cognition. These are also considered as neurodegenerative diseases that highly impacts cognitive skills. Central gray matte analysis is inevitable to track brain activity. Complete cure of mental disorder is achieved if the early medication is prognised. Hence, there has been an interest in making computational models that can help psychiatrists to detect mental disorders in their pre-stages of it. However, there is a significant gap in improving prediction that can aid psychiatrists in better prognosis. The state-of-the-art models performed binary classification, as far the authors knowledge goes, multiclass classification of psychiatric disorders are not yet performed. To address these challenges, we propose a novel model, PsychNet that can classify and track down the psychiatric diseases in the early stages. PsychNet is built with three modules. In the first, the binary classification is performed to diversify Alzheimer's and non-Alzheimer's images. The Psy-chN et enhanced classification and detection accuracy through fine tuning. Second, a novel feature extraction along with noise removal techniques are proposed using the RogerHat filters for the multi-class classification of the Alzheimer's disease types. A novel feature extraction technique is proposed to classify because of the nonlinear high complex neuroimaging data. The model's explainability is carried out in the final module by detecting the diseased area using automated features and generating a bounding box around the affected area. PsychNet surpasses the existing models to obtain classification and detection accuracies of 92% and 94%, respectively, on the dataset of fMRI scan images. The proposed model achieved 93.72 average area under the receiver operating characteristic curve (AUCROC) for balanced diseased datasets using l0-fold cross-validation. The same model architecture can also be used to detect mental disorders over the genotypes and EEG signals.