Data-driven evaluation of functional connectivity metrics

One essential problem in functional brain network study is measuring the functional connectivity among brain regions of interest (ROIs). Several widely-used functional connectivity metrics have been proposed so far in the field. However, their advantages and potential pitfalls have not been adequately examined. In this paper, we address this problem via a data-driven strategy. We perform classification experiments based on the large-scale functional connectivity patterns derived from resting-state fMRI (rs-fMRI) data and natural stimulus fMRI data (N-fMRI) of video watching, respectively. Functional connectivities were measured via commonly used metrics including the Pearson correlation (PeCo), partial correlation (PaCo), mutual information (MI), and wavelet transform coherence (WTC). The accuracies in classification tasks are then used as the criteria to evaluate the aforementioned four metrics. Our experimental results show that WTC can achieve the best classification performance in both patient-control and video classification tasks, suggesting that WTC is a preferable functional connectivity metric for functional brain network study, in at least classification applications.