Towards repeatable and converging methods in diffusion MRI: Evidence from a longitudinal chronic pain cohort

IntroductionStudies of white matter (WM) alterations in chronic low back pain (CLBP) using diffusion MRI (dMRI) have yielded inconsistent results, highlighting a need for more reproducible methods. This study introduces a longitudinal analysis pipeline designed to identify stable and repeatable WM microstructural differences between CLBP patients and healthy controls. MethodsDiffusion MRI was acquired from 27 CLBP patients and 25 control participants at three separate visits with two-month intervals. A customized Tract-Based Spatial Statistics (TBSS) analysis was performed on fractional anisotropy (FA) maps that were averaged across visits for each participant. Resulting clusters showing group differences were subsequently filtered based on cluster size (>10 voxels) and a high repeatability threshold (image intra-class correlation coefficient [I2C2] > 0.75) to ensure findings were stable across all three imaging sessions. Results were compared against standard single-visit analyses and an alternative tractometry analysis. ResultsThe repeatability analysis identified 11 clusters with stable and significant FA differences. Seven clusters, located primarily in the occipital, parietal, and frontal lobes, showed higher FA in controls. Four clusters, located in the frontal and temporal lobes, showed higher FA in the CLBP group. In contrast, single-visit analyses identified a much larger number of clusters (21 to 36), the majority of which were not spatially consistent across time and did not overlap with the final repeatable clusters. The repeatable TBSS findings demonstrated a strong spatial correspondence with group differences found using a tractometry analysis. DiscussionBy incorporating a longitudinal design and explicit repeatability filtering, our method effectively reduces spurious findings common in single-visit dMRI studies. This approach successfully isolated reliable WM regions with altered dMRI metrics in CLBP, demonstrating its value in improving the robustness of neuroimaging research in chronic pain.