Homology and topology based metrics for evaluating cortical parcellations generated using diffusion MRI

When using diffusion MRI for segmenting the cerebral cortex, the modality of information used and workflow procedural factors can have significant effects on the resulting parcellation. There is as yet no consensus on best practice processing protocols, and no ground truth is available in vivo. Converging indirect evidence has been used to compare parcellation outcomes, including: (1) comparison of cortical parcellations based on different modalities; (2) reproducibility across independent acquisitions; (3) consistency across modality or subject; and (4) the extent to which the segmented regions are functionally distinct based on task or rsfMRI data. To these we add an additional strategy wherein parcellation results are assessed based on known organizational principles of the brain, specifically inter-hemispheric homology and topology, thereby permitting assessment of results per subject independently of another imaging modality or acquisition. We propose these measures to guide improvements in acquisition, reconstruction, and/or clustering approaches during the process of diffusion MRI parcellation.