Hierarchical Atlas Organization and Inference for MRA-free Full Intracranial Vessel Geometry Mapping

Motivation: 3D MR vessel wall imaging is valuable to quantify plaque burden. Current protocols require a companion angiography image to localize and define vessel to guide subsequent VWI analysis, leading to additional effort, time, and risk of misalignment to VWI. Goal(s): To develop an MRA-free intracranial vessel mapping method. Approach: We define distance between VWI samples based on deformation complexity. K-medoid clustering is performed on a VWI-MRA atlas and the medoids are used to represent each cluster. Vessel trees is inferred from new VWI by deforming the closest medoid MRA. Results: Our method manage to generate vessel tree with high agreement with MRA. Impact: By organizing atlas with clustering, our approach provides a stable approach to infer vessel from VWI alone, mainlining the intrinsic consistency. This progress would simplify and expedite VWI protocol, and improve consistency.