Hypervoxels: a multidimensional framework for the representation and analysis of neuroimaging data

Abstract

AbstractMost neuroimaging modalities use regular grids of voxels to represent the three-dimensional space occupied by the brain. However, a regular 3D voxel grid does not reflect the anatomical and topological complexity represented by the brain’s white matter connections. In contrast, tractography reconstructions based on diffusion MRI provide a closer characterisation of the white matter pathways followed by the neuronal fibres interconnecting different brain regions. In this work, we introduce hypervoxels as a new methodological framework that combines the spatial encoding capabilities of multidimensional voxels with the anatomical and topological information found in tractography data. We provide a detailed description of the framework and evaluate the benefits of using hypervoxels by carrying out comparative voxel and hypervoxel cluster inference analyses on diffusion MRI data from a neuroimaging study on amyotrophic lateral sclerosis (ALS). Compared to the voxel analyses, the use of hypervoxels improved the detection of effects of interest in the data in terms of statistical significance levels and spatial distribution across white matter regions known to be affected in ALS. In these regions, the hypervoxel results also identified specific white matter pathways that resolve the anatomical ambiguity otherwise observed in the results from the voxel analyses. The observed increase in sensitivity and specificity can be explained by the superior ability of hypervoxel-based methods to represent and disentangle the anatomical overlap of white matter connections. Based on this premise, we expect that the use of hypervoxels should improve the analysis of neuroimaging data when the effects of interest under investigation are expected to be aligned along distinct but potentially overlapping white matter pathways.