Robust brain complexity measurement using T2* informed multivariate sample entropy from multi-echo BOLD fMRI

Abstract

Sample entropy is a useful tool to analyze resting-state fMRI data and evaluate the complexity of neural activity. However, sample entropy computed from conventional BOLD signals is not highly resistant to artifacts. In contrast, T2* may be more sensitive to certain neural-related changes, and directly incorporating T2* information enables the detection of features that are often lost in BOLD-only sample entropy due to noise influences. To further improve the robustness and reliability of complexity estimation, we introduce a multivariate sample entropy approach that incorporates T2* signals derived from multi-echo BOLD signal. We applied this method to compare a cohort of 23 elder adults and 12 younger adults, and observed significant group differences in several brain regions. Notably, the proposed method yielded more pronounced group differences than conventional sample entropy computed without T2* information. Moreover, in regions with reduced neural activity due to ischemia, as well as in ventricular areas, our approach demonstrated superior performance. These results indicate that the proposed method enhances the ability to characterize the complexity of neural activity.