A modified CNN model for diagnosing schizophrenia disease using fMRI data

Schizophrenia is a debilitating and persistent mental disorder. It manifests as cognitive, emotional, and behavioral limitations with various degrees of social function impairment that get worse with time. To identify those limitations, a deep learning-based neuroimaging technique is feasible to detect. In the last two decades, neuroimaging, precisely functional magnetic resonance imaging (fMRI) has become indispensable for investigating the brain image processing and related abnormalities of schizophrenia patients. We proposed a modified Convolutional Neural Network (CNN) model based on VGG16. Along with our proposed model, a set of CNN models—VGG16, VGG19, EfficientNet, Inception V3, and ResNet-50 have been considered to identify schizophrenia and healthy controls. We evaluated correlation analysis on time series feature extractions over Open-NEURO clinical dataset and achieved accuracy of schizophrenia detection of 99.5%. Moreover, Receiver Operating characteristic Curve (ROC) have been generated to compare with existing reference models. This research significantly affects the primary detection of schizophrenia and might prevent a patient from mental disorder at an early stage.