Hypervoxels: A Multidimensional Framework For The Representation And Analysis Of Neuroimaging Data

Most neuroimaging modalities use regular grids of voxels to represent the three-dimensional space occupied by the brain. However, the topological uniformity of the 3D voxel grid does not realistically reflect relevant properties of the brain's anatomy such as its heterogeneity and the complexity of its structural connectivity. In contrast, tractography reconstructions based on diffusion MRI can provide compelling characterisation of the brain's structural connectivity that reflects the local and global organisation of the neuronal fibres. In this work we introduce hypervoxels as a new analysis framework that combine the anatomical and connectivity information provided by tractography streamlines with the spatial encoding capabilities of multidimensional voxels. We provide a detailed description and implementation of the framework and demonstrate its benefits by carrying out comparative voxel and hypervoxel analyses on neuroimaging data acquired to investigate the neuroanatomical substrate of motor neurone disease (MND). The hypervoxel analyses outperformed the corresponding voxel analyses in their ability to detect effects of interest in brain regions known to be affected in MND, as demonstrated by the widespread number of significant results, their statistical significance and their anatomical specificity. We conclude that the anatomical and topological information provided by the hypervoxel framework can greatly improve the sensitivity and anatomical specificity of neuroimaging analyses through its application to any method or imaging modality based on traditional voxels. ### Competing Interest Statement The authors have declared no competing interest.