Resting-state functional magnetic resonance imaging (R-fMRI) has been widely used in the field of basic research and clinical research for decades due to its simply experimental design, timesaving acquisition, nonradiative resonance and high spatial resolution. However, noise reduction, replicability and standard pipeline still threat the further development of R-fMRI. Global signal regression (GSR) has been the most controversial step during the preprocessing of R-fMRI data. Whether or not using GSR has a considerable impact on the analysis results of R-fMRI for it may reduce noise of R-fMRI and leads to the zero-centered of data. In order to explore the effect of use of GSR and characterizing physiology in human brain, we utilize a large sample dataset with high quality and acquisition of physiological signal. Data preprocessing and analysis was performed with highly selected pipeline and statistical methods. Analysis focused on the physiology of human brain, noise reduction of GSR and how GSR influence R-fMRI analysis.
Previous studies indicated the complex mechanism of physiology noise in human brain includes head motion, heartbeat and respiratory which weaken the replicability and validity of R-fMRI research. Nevertheless, no empirical study has characterized dynamic physiological network by sliding window and verified the GSR in reduction of physiological noise. Thus, study 1 tried to demonstrate physiological noise and evaluate the effect of GSR on potential strength. Study 1 revealed a large widespread physiological network in human brain and the application of GSR may reduce the contaminate of head motion.
Recently, replicability has been a major concern in neuroscience. The lacking of standard pipeline and complex noise undermine the findings of R-fMRI research. GSR is a critical problem since it leads to different conclusion even in the same dataset. Hence, study 2 used dataset with two session to examine the replicability of physiological noise and traditional R-fMRI analysis after the preprocessing of GSR. Results of study finds the high replicability of physiological noise, while GSR may decrease the replicability of seed-based functional connectivity and brain network analysis. In five indices, the replicability of amplitude of low frequency fluctuations (ALFF) and fractional ALFF (fALFF) had little difference both with and without GSR. However, the replicability of regional homogeneity (ReHo), voxel-mirrored homotopic connectivity (VMHC) and degree centrality (DC), had been substantially decreased after use of GSR.
Each subject was used the same pipeline in the R-fMRI which neglected the difference of individual noise. Some subject with more noise required further noise reduction while those with little noise may not need. Hence, study 3 used canonical correlation analysis (CCA) to discriminate subjects with notorious noise and with little noise by explore of relationship between R-fMRI signal and physiological noise. Then based on the results of CCA, performing GSR or not was determined. Study 3 indicated that head motion plays a major part in R-fMRI signal and other noise may weighted little in physiological variables.
In summary, the present study demonstrated a wide physiological noise network in human brain and indicated that using GSR may mitigate the effect of head motion on R-fMRI data; after the preprocessing of GSR, the replicability of seed-based functional connectivity and brain network decreased; the preprocessing of GSR had little impact on the replicability of ALFF and fALFF, while largely impaired the replicability of other indices; CCA results revealed that head motion is the most significant factor among those physiological noise and GSR may help to denoise. The present study found the connection between physiological noise and R-fMRI signal. The use of GSR may decrease the replicability of major R-fMRI analysis results. Using CCA to determine the use of GSR had a fundamental step for individual preprocessing and understanding the complexity of physiological network.
修改评论