Distinctive BOLD Connectivity Patterns in the Schizophrenia Brain: Machine-learning based comparison between various connectivity measures

Abstract

Recent functional magnetic resonance imaging (fMRI) studies have found distinctive functional connectivity in the schizophrenic brain. However, most of the studies focused on the correlation analysis to define the functional connectivity for BOLD fluctuations between brain regions, which resulted in the limited understanding to the network properties of altered wirings in the schizophrenic brain. Here I characterized the distinctiveness of BOLD connectivity pattern in the schizophrenic brain relative to healthy brain with various similarity measures in the time-frequency domain, while participants are performing the working memory task in the MRI scanner. To assess the distinctiveness of the connectivity pattern, discrimination performances of the pattern classifier machines trained by each similarity measurement were compared. As a result, the classifier machine trained by cross-spectral coherence pattern in low frequency fluctuation (LFF, 0.01-0.08 Hz band) made better performance than the machine trained by correlation-based connectivity pattern. Interestingly, the classifier machine trained by time-lagging patterns in cross-correlation function of LFF produced higher classifying sensitivity than the machines trained by other measures. These results indicate that there are unobserved but considerable features in the functional connectivity pattern of schizophrenic brain which traditional emphasis on correlation analysis does not capture.