Clinical Fellow Brigham and Women's Hospital Boston, MA, US
Introduction: Understanding topological symmetry of whole-brain connectomes is vital to neuroscience and neurosurgery. Current methodologies for brain parcellation involve registering brain atlases to individuals’ brains. However, despite alignment of atlases to person's brain, these methods lack specificity needed for individual variations. In order to overcome this, we analyzed topological symmetry of whole-brain connectomes and functional subnetworks in healthy subjects and diffuse astrocytomas (DA) patients. Relying on interhemispheric symmetry assumption, we developed novel personalized parcellation methodology that maximizes connectomes’ topological symmetry.
Methods: DTI/T1 images of 10 subjects from HCP-MMP dataset and from 10 DA patients acquired at Hacettepe University. We first constructed whole-brain connectomes and functional subnetworks for these subjects using MRtrix3. Subsequently, we quantified topological symmetries between left and right hemisphere networks using the Jaccard Index. Finally, we proposed new mathematical optimization model for personalized parcellation that re-assigns voxels to parcels, aiming to maximize topological symmetry while allowing only limited deviation from baseline parcellation.
Results: Jaccard Indices averaged across 10 HCP-MMP subjects quantifying topological symmetry of connectomes obtained with baseline parcellation were as follows: (0.355, 0.637, 0.631, 0.588, 0.764) for Whole-Brain, Default-Mode, Central-Executive, Language, Auditory networks, respectively. After running our parcellation model, Jaccard Indices improved significantly: (0.771, 0.947, 0.820, 0.890, 0.963), respectively. Moreover, the similarity of personalized parcellations to atlas-based parcellations was maintained at 0.812. Analysis for DA cases yielded the following Jaccard indices: (0.302, 0.567, 0.654, 0.450, 0.526). Findings in DA patients suggest that connectomes created using atlas-based parcellation also lack coherent topological symmetry. Compared to healthy subjects, the impact of the tumor likely results in more reduced symmetry.
Conclusion : The proposed model achieved personalized parcellation in healthy individuals by improving topological symmetry. Tumor patients’ preliminary analysis indicates that improvements may achieved by a model similar to that applied over healthy subjects, this time also considering the effects of the tumor.
.jpg)