Introduction

Multi-site diffusion imaging studies are increasingly used to study brain disorders. However, inter-site and inter-scanner variability in the acquired data sets poses a potential problem for joint analysis of diffusion MRI (dMRI) data. It was deemed that there exist large differences between diffusion measurements from different sites. Specifically, the inter-site variability in FA and MD is not uniform over the entire brain, but is tissue specific as well as region specific. Inter-site variability in FA can be up to 5% in major white matter tracts[1]. This study aimed to confirm the variability and to harmonize data across sites.

Methods

5 healthy young adults (male; age range, 21~23 years) participated in the MR imaging of brain. Each participant underwent one DTI scan in four sites, respectively. This study was approved by the local Institutional Review Board. All parents of participants were informed with the goals and risks of MR scanning and requested written consent before enrollment. In this study, all the scans were performed using 3T MR scanners (site1: MR 750, GE Healthcare; site2: Ingenia, Philips; site3: skyra, Siemens; site4: MR 750, GE Healthcare). Other acquisition parameters were shown in Table 1. The DTI data were automated processed in MATLAB toolbox named “Pipeline for Analyzing brain Diffusion images” ( PANDA). The process mainly included extracting brain images, correcting eddy current, calculating the DTI metrics(FA,MD,AD,RD) and tract-based spatial statistics (TBSS)[2]. We quantified the diffusion MRI variability across sites using reproducibility examined by measures of variability (absolute error relative to the sub and sum about the two scans) across subjects and sites. Then, we harmonized DTI data acquired on scanners of three different manufacturers repeated scanning of 5 participants on GE (site1, site4), Siemens (site2) and PHILIPS (site3) scanners[3]. Specifically, we regarded site1 as the reference group. Then, we inferred a scanner-specific correction factor from the skeletons in different sites and applied this correction factor to the DTI-derived maps in site2, site3, site4. Specifically, skeletons were inferred from the manufacturer-corrected FA maps through TBSS, which enabled a quantitative survey of white matter of the entire brain. In parallel, permutation testing was applied to the DTI metrics derived skeletons.

Results

Five male young adults with mean age of 21.8±0.8 of 110.2±9.0 (age range, 21~23 years) were included in this study. The comparisons of FA reproducibility among manufacturers were shown in Fig.1. The results showed significant differences. For the parameters reproducibility between site1 and site4, there were 7% voxels in the skeletons whose FA reproducibility were greater than 0.2, which was less than 12% (site1 vs site2) and 26%(site1 vs site3). The results of other DTI metrics were shown in Table 2. We therefore corrected each map by applying the ratio between two sites with respect to the mean values across the 5 participants: Mean FA site1/ Mean FA site2=0.89; Mean FA site1/ Mean FA site3=0.911; Mean FA site1/ Mean FA site4=0.93. After correction, 3.3% voxels were found that FA reproducibility of that in the skeletons were greater than 0.2 when site1 was compared to site4, while we found that there were 6% in site1 vs site2, and 7.4% in site1 vs site3. Original and corrected means of DTI skeletons for each subjects were shown in Fig.2.

Discussion

This study demonstrated the inconsistency of DTI metrics measured. Among four sites, and further suggested the significance of data harmonization across MRI sites. Being consistent with previous study, our results showed that there existed inter-site and inter-scanner variability for joint analysis of dMRI data. Besides, we also can see a smaller variation between sites with the same manufacturer, so it also should be corrected since the nonnegligible variation. Furthermore, to reduce variation within the DTI-based measurements unrelated to factors exclusive of the primary study hypotheses, we applied a harmonization approach to the multi-site data. We obtained good consistency with a good extent (96.7% between site1 and site4; 94% between site1 and site2; 92.6% between site1 and site3 in the skeletons with a FA reproducibility smaller than 0.2. Besides, After correction, mean DTI metrics in the skeletons in site2, site3, site4 substantially drew near to the reference site. Future study is desired to perform a large data-driven robust model to harmonize the DTI data across sites in multicenter study.

Conclusion

The variability of DTI metrics in the human brain is obvious across various sites. Harmonizing DTI measurements across scanners would be considered as an effective method to improve the variability across sites and across scanners.

Acknowledgements

This study was supported by the National Natural Science Foundation of China (Project Nos. 81371630, 81571752, the Shaanxi Nova program, the Fundamental Research Funds for the Central Universities, National key research and development plan of China, 2016YFC0100300.