Combining magnetic resonance imaging (MRI) data from multi-site studies is a popular approach for constructing larger datasets to greatly enhance the reliability and reproducibility of neuroscience research. However, the scanner/site variability is a significant confound that complicates the interpretation of the results, so effective and complete removal of the scanner/site variability is necessary to realize the full advantages of pooling multi-site datasets. Independent component analysis (ICA) and general linear model (GLM) based harmonization methods are the two primary methods used to eliminate scanner/site-related effects. Unfortunately, there are challenges with both ICA-based and GLM-based harmonization methods to remove site effects completely when the signals of interest and scanner/site-related variables are correlated, which may occur in neuroscience studies. In this study, we propose an effective and powerful harmonization strategy that implements dual-projection (DP) theory based on ICA to remove the scanner/site-related effects more completely. This method can separate the signal effects correlated with site variables from the identified site-related effects for removal without losing signals of interest. Both simulations and vivo structural MRI datasets, including a dataset from Autism Brain Imaging Data Exchange II and a traveling subject dataset from the Strategic Research Program for Brain Sciences, were used to test the performance of DP-based ICA harmonization method. Results show that DP-based ICA harmonization has superior performance for removing site effects and enhancing the sensitivity to detect signal of interest as compared with GLM-based and conventional ICA harmonization methods.

Static features of the executive control network (ECN), dorsal attention network (DAN), default mode network (DMN), and salience network (SN) have displayed dysfunction in people with autism spectrum disorder (ASD). However, research on the dynamic brain function of these networks in ASD is rare. In this study, co-activation pattern (CAP) analysis was performed on the whole cortex to study dynamic dysfunction in ASD using a large multisite resting-state fMRI dataset (295 ASDs, 446 healthy controls). Eight transient network states (TNSs) were defined, the dwell time, persistence, and transitions of each TNS were calculated to evaluate dynamic brain function. Using hierarchical clustering, the eight TNSs were divided into three clusters: ‘DMN activating’, ‘SN activating’, and ‘ECN and DAN activating’. We found ‘ASD-biased’ DMN and SN TNSs, which showed larger dwell time and longer persistence in ASD group than healthy control (CON) group. More transition within ‘ASD-biased’ TNSs were found in ASD group. Dwell time of the ‘ASD-biased’ ‘SN activating’ TNS was significantly correlated with social deficits only in the ASD group. Our results imply the dynamic dysfunction of ASD does not come from the occurrence of DMN, ECN, or SN, but comes from the atypical co-activation patterns within them. Our results also indicate people with ASD have stronger negative connectivity between DMN and ECN in childhood. This connection dosen’t change significantly with age in ASD group, but is supposed to increase with age until adulthood as the growth trajectory in healthy inviduals, which implies the early overgrowth of ASD children.