Persistent Homology Distances for Comparing Disease-Filtered Structural Connectomes

Hamza Tahir Chaudhry, Vineet R Tiruvadi, Cengiz Pehlevan, Michael D Fox MD PhD, Kanaka Rajan

Keywords: Computational Psychiatry, Structural Connectomics, Diffusion MRI, Deep Brain Stimulation, Persistent Homology

TL;DR: We propose a persistent homology–based metric to compare disease-specific connectomes, revealing consistent hemisphere < disease < site differences and offering an objective tool for connectome evaluation and DBS applications.

Abstract: Whole-brain circuit mapping is yielding wiring diagrams of axon bundles, or connectomes, that improve both scientific understanding and clinical care of brain disorders. However, the lack of direct, objective metrics to compare disease-relevant connectomes potentially hinders progress. Here, we introduce early steps towards building such metrics with parcellation-free, geometry-aware method for analyzing disease-specific streamline bundles using persistent homology and distributions within bundles. We use persistent homology for multiscale comparison of streamlines in bundles with Wasserstein distance distributions to characterize topological distances in disease-specific, or filtered, connectomes. We observed heirarchical similarity hemisphere < disease < connectome. This measure may serve as a foundation to guide costly next-generation connectome generation while also optimizing disease-specific connectome atlases for neural implant design, implantation, and programming.

Submission Number: 67