Participants and Imaging Protocols
MRI data were obtained from the Consortium for Reliability and Reproducibility (CoRR)21. The 36 datasets from the CoRR originally included 1725 participants who underwent at least two scanning sessions. Here, we only chose the baseline R-fMRI data to analyse the functional connectivity difference between males and females. With the above exclusion criteria, we made quality control. From this perspective, 1291 participants from 30 datasets were selected (age 25.796 ± 15.521, 671 females, see Table 1 for details). Six datasets were excluded due to the loss of information or the small sample size of participants. Participants from the rest of the datasets were excluded if their head motion was excessive (more than 2.5 mm of maximal translation in any direction of x, y, or z or 2.5°of maximal rotation throughout scanning). To control the confounds of handedness, participants with non-right handedness were excluded. Participants with low-quality normalization or incomplete brain coverage were excluded. In addition, age was matched (p > 0.5, two sample T-test) between male and female groups in each dataset.
Data Preprocessing
All preprocessing of resting-state fMRI data was processed using RESTplus V1.2425. Initially, the first 10 time points were discarded to overcome the influence of instability when the scanner was switched on while participants adapted to the scanner's noise. Second, a slice-timing correction was performed to correct the acquisition time delay for all volumes. Third, head motion correction. Fourth, individual structural images were co-registered to mean functional images, then, the co-registered structural images were segmented into gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), bone, soft tissue and air/background26. Parameters generated from step 4 were used to apply for functional images spatially normalized to Montreal Neurologic Institute space (the resampling voxel size = 3 mm × 3 mm × 3 mm). Fifth, normalized fMRI data were smoothed with a Gaussian kernel of 6 mm × 6mm × 6mm full-width at half maximum (FWHM). Sixth, the linear trend of the time course was removed. Next, head motion effects (using Friston 24 parameters) were regressed out to minimize head motion confounds. Finally, data were temporally band-pass filtered (0.01Hz – 0.08Hz).
Functional Connectivity Calculation
Fifty-nine coordinates within DMN were extracted from the 11 studies, and all coordinates were converted to Montreal Neurological Institute space. As for resting-state functional connectivity, we defined fifty-nine spherical regions of interests (ROIs) with a radius of 6 mm centered on the fifty-nine coordinates. The radius was defined in line with a prior study14. After data preprocessing, the time course of each seed ROI was extracted, and functional connectivity was calculated by computing the Pearson correlation coefficient between the mean signal time course from the ROI and all other voxels in the entire brain.
Statistical Analysis
To increase the normality of the distribution of correlation, all FC maps were transformed through Fisher’s r-to-z transformation. Further, two-sample t-tests were performed using Data Processing & Analysis for Brain Imaging (DPABI)27 to compare the difference between males and females' FC maps, which Fisher’s r-to-z transformation has processed. Finally, all t maps were used for subsequent image-based meta-analysis.
Literature search
A literature search of relevant articles was conducted in Web of Science and PubMed as of December 13, 2020, using the keywords “default mode network” or “DMN”. First, the titles of highly cited papers on the topic labeled by the databases were recorded. Then, a full-text search for these articles was performed to assess the documents for those that provided coordinates of DMN nodes. Finally, the reference lists of articles with coordinates of DMN nodes were manually scanned for other articles that were not retrieved at the database search phase. When multiple articles reported exact coordinates, the earliest one was included. And if the coordinates reported in the papers were quoted, the source article was selected instead. When the coordinates of the same article were inconsistent in different references, the coordinates reported in the original research were included. As a result, 68 articles were identified by our search. These criteria generated 59 coordinates of DMN nodes from 11 articles that were retrieved from our searches.
Data Extraction and Coding
All coordinates were converted to Montreal Neurological Institute space using the tal2icbm transform28. According to the converted coordinates, each seed was categorized into a brain region of anatomical automatic labeling (AAL) atlas29 (see Table 2 for details). Visualization of seeds and brain network were shown through the BrainNet Viewer software30.
SDM Meta-Analysis
We performed an image-based meta-analysis named Anisotropic Effect-Size Signed Difference Mapping (AES-SDM, version 5.15) to examine the FC differences between male and female. The AES-SDM approach uses full statistical images as input and allows both positive and negative values of the same map to be preserved (Joaquim Radua et al., 2014). This approach has been found valid and well described in previous studies31–34.
For each center, AES-SDM constructs an effect size and corresponding variance map from the unthresholded T-map. The method uses a standard random-effects model, which considers sample size, study precision and between-study heterogeneity, to combine the maps of different effect sizes and variance of different centers. The random-effects model ensures that studies with larger sample size or lower variability contribute more to the meta-analysis results. A randomization test that randomizes the location of the voxels within the SDM gray matter template was performed to assess the statistical significance. Since the corrected P value in fMRI articles is seriously affected by methodological factors, AES-SDM uses a combination of two thresholds, in which the uncorrected p = 0.005 is selected as the main threshold. In addition, a Z-based threshold is added to reduce the possibility of false-positive results (we set z > 1.00, which is the default setting of AES-SDM). Meanwhile, 10 voxels were applied to threshold the cluster size.