Schizophrenia Classification based on Structural and Functional MRI using 3D Convolutional Neural Networks with Multimodal Approach

A serious mental illness, schizophrenia (SZ) usually manifests in early adulthood. SZ is a chronic, complicated mental illness that results in abnormal behavior in its victims and is defined by a variety of symptoms that affect reality perception. The automatic prediction of an individual’s diagnosis of schizophrenia has been suggested to be facilitated by Magnetic Resonance Imaging (MRI). To improve accuracy, ongoing research is being done on the best ways to integrate data from various MRI modalities. The approach involved combining structural and functional MRI data into a multimodal classification paradigm to benefit from their complimentary knowledge. Our effort focuses on developing an adaptive control unit for the fusion phase and expanding neural networks for feature learning. By applying 3D-Convolutional Neural Networks (3D-CNN) and Fast-ICA with a 10-fold cross-validation method, the suggested model outperformed several existing models for unimodal or multimodal data, achieving 99.3% accuracy in schizophrenia identification.