Detecting Functional Connectivity in fMRI Using PCA and Regression Analysis

Abstract

A fMRI connectivity analysis approach combining principal component analysis (PCA) and regression analysis is proposed to detect functional connectivity between the brain regions. By first using PCA to identify clusters within the vectors of fMRI time series, more energy and information features in the signal can be maintained than using averaged values from brain regions of interest. Then, regression analysis can be applied to the extracted principal components in order to further investigate functional connectivity. Finally, t-test is applied and the patterns with t-values lager than a threshold are considered as functional connectivity mappings. The validity and reliability of the presented method were demonstrated with both simulated data and human fMRI data obtained during behavioral task and resting state. Compared to the conventional functional connectivity methods such as average signal based correlation analysis, independent component analysis (ICA) and PCA, the proposed method achieves competitive performance with greater accuracy and true positive rate (TPR). Furthermore, the 'default mode' and motor network results of resting-state fMRI data indicate that using PCA may improve upon application of existing regression analysis methods in study of human brain functional connectivity.