Abstract
Previous resting-state functional magnetic resonance imaging (rs-fMRI) studies frequently applied the spatial normalization on fMRI time series before the calculation of temporal features (here referred to as "Prenorm"). We hypothesized that calculating the rs-fMRI features, for example, functional connectivity (FC), regional homogeneity (ReHo), or amplitude of low-frequency fluctuation (ALFF) in individual space, before the spatial normalization (referred to as "Postnorm") can be an improvement to avoid artifacts and increase the results' reliability. We utilized two datasets: (1) simulated images where temporal signal-to-noise ratio (tSNR) is kept a constant and (2) an empirical fMRI dataset with 50 healthy young subjects. For simulated images, the tSNR is constant as generated in individual space but increased after Prenorm and intersubject variability of tSNR was induced. In contrast, tSNR was kept constant after Postnorm. Consistently, for empirical images, higher tSNR, ReHo, and FC (default mode network, seed in precuneus) and lower ALFF were found after Prenorm compared to those of Postnorm. Coefficient of variability of tSNR and ALFF was higher after Prenorm compared to those of Postnorm. Moreover, the significant correlation was found between simulated tSNR after Prenorm and empirical tSNR, ALFF, and ReHo after Prenorm, indicating algorithmic variation in empirical rs-fMRI features. Furthermore, comparing to Prenorm, ALFF and ReHo showed higher intraclass correlation coefficients between two serial scans after Postnorm. Our results indicated that Prenorm may induce algorithmic intersubject variability on tSNR and reduce its reliability, which also significantly affected ALFF and ReHo. We suggest using Postnorm instead of Prenorm for future rs-fMRI studies using ALFF/ReHo.