General Microstructure Factor Analysis of Diffusion MRI in Gray-Matter Predicts Cognitive Scores

TL;DR

The study investigates whether global gray-matter diffusion microstructure patterns relate to cognition by deriving a general gray-matter microstructure factor (GMF) from region-averaged NODDI parameters across 68 cortical regions in the HCP-YA cohort using PCA. The first GMF, particularly for the isotropic volume fraction $f_{iso}$, explains a substantial portion of regional variance ($40.4\%$) and shows significant correlations with NIH Toolbox reading and vocabulary scores after controlling for partial-volume effects with a weighted-average volume covariate. Region-level analyses corroborate a broad structure–function link, with 54 regions associated with ReadEng and 59 with PicVocab, and 26 with CardSort. Overall, GMFs emerge as robust, interpretable global biomarkers of cortical organization that complement region-specific diffusion analyses for understanding cognition.

Abstract

Diffusion MRI has revealed important insights into white matter microstructure, but its application to gray matter remains comparatively less explored. Here, we investigate whether global patterns of gray-matter microstructure can be captured through neurite orientation dispersion and density imaging (NODDI) and whether such patterns are predictive of cognitive performance. Our findings demonstrate that PCA-based global indicators of gray-matter microstructure provide complementary markers of structure-function relationships, extending beyond region-specific analyses. Our results suggest that general microstructure factors may serve as robust, interpretable biomarkers for studying cognition and cortical organization at the population level. Using diffusion MRI and behavioral data from the Human Connectome Project Young Adult study, we derived region-averaged NODDI parameters and applied principal component analysis (PCA) to construct general gray-matter microstructure factors. We found that the factor derived from isotropic volume fraction explained substantial inter-individual variability and was significantly correlated with specific cognitive scores collected from the NIH Toolbox. In particular, the isotropic volume fraction factor was linked to reading and vocabulary performance and to cognitive fluidity.

Figures (5)

• Figure 1: We study the relationship between global trends in NODDI gray-matter microstructure estimates and cognition using the procedure outlined by the figure. For each subject, we construct a vector with entries equal to per-region mean microstructure estimates. This produces, for each subject, a vector with one entry per cortical region. We then use PCA analysis to compute a general microstructure factor (GMF) which assigns weights to each cortical region corresponding to that region's participation in global variation in microstructure parameter. Each subject is thus assigned a GMF value. We then study trends between GMF and cognition using NIH Toolbox cognitive scores, and find a significant correlation between GMF and three cognitive scores.
• Figure 2: Distribution of mean per-region microstrucutural values across subjects. Extremal values corresponding to misregistration or distortion artifacts have been filtered. This is an intermediate result; principal component analysis is performed on these distributions to produce the general factors with weights pictured in Fig. \ref{['fig:pca-weights']}.
• Figure 3: General factor weights corresponding to the first principal components of the distributions pictured in Fig. \ref{['fig:params-per-reg']}. The height of each bar corresponds to the entry of eigenvector associated with that region, i.e., how much variation in microstructure for that region contributes to global variation.
• Figure 4: Scree plots for the PCA analysis of the mean microstructural parameters for each region, Fig. \ref{['fig:params-per-reg']}.
• Figure 5: Lines of best fit for each of the significant trends observed. The dependent axis plots the general microstructure factor (GMF) of the corresponding NODDI parameter.