An interpretable approach for schizophrenia classification using fMRI and sMRI features

Comprehensive Summary

In this study, researchers developed a novel framework for the classification of schizophrenia by integrating structural MRI (sMRI) and functional MRI (fMRI) data through a Multi-scale Recurrent Neural Network (MsRNN). The model utilizes Independent Component Analysis (ICA) to extract 410 neuroimaging features, which were broken down into 378 Functional Network Connectivity (FNC) and 32 Source-Based Morphometry (SBM) features. These were used to capture both dynamic brain activity and physical structural alterations. Validated on the MLSP 2014 and COBRE datasets, the MsRNN achieved high classification accuracies of 83.33% and 89.8%, respectively, outperforming traditional machine learning models like Random Forest and SVM. To address the "black box" nature of deep learning, the framework integrates Explainable AI (XAI) techniques, specifically Layer-wise Relevance Propagation (LRP) and Gradient-weighted Class Activation Mapping (Grad-CAM), to provide visual heatmaps of the brain regions most influential in the model's decision

Outcomes and Implications

The clinical implications of this research are significant, as it offers a more transparent and objective alternative to traditional diagnostic methods that often rely on subjective behavioral assessments. By identifying specific neurobiological markers, such as gray matter reduction in the frontal and temporal lobes or hyperactivation in the prefrontal cortex, the framework provides clinicians with evidence-based visualizations to support diagnostic decisions. The use of XAI specifically fosters clinical trust, allowing medical professionals to see exactly which anatomical regions, such as enlarged ventricles or altered connectivity in the default mode network, are driving a diagnosis. This bridge between high-performance machine learning and interpretable clinical insights could ultimately facilitate earlier interventions and the development of personalized treatment strategies for individuals with schizophrenia