Re-analysis of MRI data shows multivariate approach detects more age-related brain changes but has cross-validation limitations

This study applied multivariate ANOVA to investigate age-related microstructural changes in the brain tissues driven primarily by myelin, iron, and water content, as observed in MRI (semi-)quantitative R1, R2*, MTsat and PD maps. This is effectively a re-analysis of the data analyzed in a univariate way by Callaghan et al., 2014. Voxel-wise analyses were performed on gray matter (GM) and white matter (WM), in addition to region of interest (ROI) analyses. The multivariate approach identified brain regions showing coordinated alterations in multiple tissue properties and demonstrated bidirectional correlations between age and all examined modalities in various brain regions, including the caudate nucleus, putamen, insula, cerebellum, lingual gyri, hippocampus, and olfactory bulb. The multivariate model was more sensitive than univariate analyses, as evidenced by detecting a larger number of significant voxels within clusters in the supplementary motor area, frontal cortex, hippocampus, amygdala, occipital cortex, and cerebellum bilaterally. Though when cross validating the results by splitting the data into 2 subsets, sensitivity is strongly reduced, even more so for the multivariate approach. The examination of normalized, smoothed, and z-transformed maps within the ROIs revealed concurrent age-dependent alterations in myelin, iron, and water content. These findings contribute to our understanding of age-related brain differences and provide insights into the underlying mechanisms of aging. The study emphasizes the importance of multivariate analysis for detecting subtle microstructural changes associated with aging when dealing with multiple quantitative MRI parameter maps.