A Novel Skeletonization Algorithm for Topologically Complex Structures: Comparative Analysis and Application to Renal Arterial Trees

Accurate skeletonization of complex vascular structures is essential for clinical applications and anatomical research. This paper presents a domain-adapted skeletonization algorithm specifically developed for the intrarenal arterial tree, with a focus on high-resolution μ-CT scans of human renal corrosion cast specimens. These datasets pose challenges due to the complexity of the vascular tree, making it difficult to convert raw μ-CT data into meaningful representations. Without proper reconstruction, skeletonization, and graph representation, raw μ-CT data remains an unstructured point cloud, unsuitable for quantitative analysis. State-of-the-art methods, including voxel coding, geometric flow, gradient vector flow, and learning-based approaches, fail to meet key performance indicators (KPIs), such as one-voxel-wide centerline extraction, topological preservation, suppression of false bifurcations, and robustness to noise. To address these limitations, the proposed method integrates vascular-specific termination criteria, branch-preserving pruning, and voxel-based interpolation to ensure anatomically accurate, topologically consistent skeletons suitable for graph-based vascular modeling. The method was quantitatively evaluated on artificial renal tree models using metrics such as branch count per hierarchical level and bifurcation angle fidelity. Qualitative assessments were performed on standardized 3D benchmark models and μ-CT scans of human renal arterial corrosion cast specimens. Results demonstrate superior performance compared to state-of-the-art techniques, particularly in preserving connectivity, minimizing spurious branches, and achieving consistent one-voxel skeleton width all while maintaining a favorable balance between accuracy and computational efficiency. The algorithm enables detailed morphological and topological analysis of renal vascular trees, supporting the identification of intra- and inter-tree structural dependencies. These capabilities contribute to improved anatomical insight and have potential applications in surgical planning, where accurate vascular mapping enhances arterial localization and patient safety. Beyond renal vasculature, the method’s adaptability makes it applicable to other tree-like anatomical systems, supporting 3D centerline extraction and structural analysis in broader clinical and research contexts. Future work will focus on integrating <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> vascular pattern knowledge into the segmentation framework to enhance performance on patient-specific μ-CT datasets, where resolution is often constrained by radiation dose limitations.