Exploring Schizophrenia Classification in fMRI Data: A Common Spatial Patterns(CSP) Approach for Enhanced Feature Extraction and Classification

In the exploration of dynamic changes in network connectivity within resting-state functional magnetic resonance imaging (rs-fMRI), the dominant focus has traditionally been on a holistic study of the entire brain. Various methodologies and analyses have been applied in prior research within this domain. This study takes a novel approach by delving into a comparative analysis of the similarities between electroencephalogram (EEG) signals with motor imagery tasks and rs-fMRI signal. Both data types collect time series data from their respective datasets. Drawing from the insights of previous research, the common spatial patterns (CSP) method, mostly used for its efficacy in handling EEG signals, was employed. Notably, CSP is a supervised learning transformation of signals, offering advantages over the implementation of deep learning models. this study pioneers the integration of the CSP method with fMRI datasets. Validation of this approach was conducted through a rs-fMRI study focused on schizophrenia, includes two primary classes: patients and controls. In addition to CSP, principal component analysis (PCA) was explored as an unsupervised dimensionality reduction technique, serving as a benchmark for comparison. The results revealed that CSP has better performance relative to PCA and other examined methods. This study contributes to the expanding landscape of understanding time-varying network connectivity, emphasizing the potential applicability of CSP beyond its traditional domain of EEG signals, and take benefit of its effectiveness in the context of rs-fMRI.