HARP: Hierarchical Anatomical Refinement of Pathways for Whole-Brain Tractography

Diffusion MRI tractography is an ill-posed inverse problem that requires anatomical constraints to ensure the plausibility of reconstructed white-matter pathways. Yet, encoding whole-brain constraints is challenging because neuroanatomical knowledge is fragmented and unevenly distributed across regions and spatial scales. We introduce HARP, a flexible framework that hierarchically injects anatomical constraints at different levels of detail, in parallel with more and more granular brain segmentations. Unlike existing fixed rule-based approaches, HARP enables the systematic, scalable integration of diverse priors within a unified framework. Across multiple tractography algorithms and acquisition protocols, HARP achieves up to a 9% reduction in implausible streamlines. These rejected connections are absent from a ground truth brain phantom, indicating their artifactual origin, and cannot be reliably identified using post-hoc filtering weights alone. By reducing false positive reconstructions, HARP improves the anatomical specificity of tract reconstructions and downstream connectome estimates. More broadly, HARP represents a step toward a collaborative effort to encode neuroanatomical insights in tractography pipelines, with the goal of advancing in vivo whole-brain connectivity studies.