Evaluating the retest reproducibility of intrinsic connectivity network using multivariate correlation coefficient

Abstract

Recently, the retest reproducibility of intrinsic connectivity networks (ICNs) has become an increasing concern in the fMRI research community. However, few indexes can be applied to directly quantify the similarity of three or more ICNs for evaluating the retest reproducibility of ICNs. To solve this problem, a multivariable correlation coefficient based on zero-mean normalization and intraclass correlation coefficient (Z-ICC) is proposed. After demonstrating the calculation method and performance analysis on theory, Z-ICC is adopted to evaluate the similarity of three ICNs from three ICN sets, which are inferred from the open retest resting-state fMRI dataset NYU_TRT with dual temporal and spatial sparse representation (DTSSR). The reproducible ICNs and quantization index of retest reproducibility are obtained by the calculated Z-ICC values and the accepted evaluation criterion. The experimental results and visual inspection show that Z-ICC can effectively identify the reproducible ICNs and quantify the retest reproducibility of ICNs. Eighteen (Z-ICC > 0.8) of the inferred twenty ICNs with DTSSR that are found to be reproducible are far more than the seven reproducible ICNs based on temporal concatenation group ICA (TC-GICA). Furthermore, the result of the one-tailed two-sample T test demonstrates that the Z-ICC values of the reproducible ICNs by DTSSR are significantly greater than those of TC-GICA, indicating that more reproducible group-level ICNs with higher retest reproducibility can be achieved with DTSSR.