Abstract
Centerline graphs are a standard representation for vascular analysis, but constructing graphs that are simultaneously geometrically accurate and topologically correct remains challenging. Such structural fidelity is an important prerequisite for downstream morphometric comparisons across populations. Existing methods generally suffer from a fundamental trade-off: tracking approaches achieve high precision but depend on ground-truth-derived initialization priors and miss large portions of the vessel tree, whereas end-to-end detectors achieve high recall but fragment connectivity across crop boundaries. To resolve this, we propose a two-stage method that decouples node localization from edge assignment and requires neither a seed nor a segmentation mask at inference. In Stage 1, a fully-convolutional detector localizes centerline nodes across the full angiogram volume using a seamless, lattice-aligned sliding window. In Stage 2, a graph neural network scores $k$-nearest-neighbor candidate edges using image features sampled from the frozen detector backbone, after which a structure-aware decoder resolves the edges into a topologically consistent graph. Evaluated on cerebral (TopCoW, MR, and CT) angiography, the proposed method achieves state-of-the-art node detection, edge connectivity, and branch completion, improving node F1 to $0.85$–$0.91$ against $0.33$–$0.71$ for three baselines, while substantially exceeding their branch-completion scores. The resulting graphs also reproduce ground-truth vascular morphometry far more faithfully, staying within a few percent of ground truth on most descriptors, providing a strong basis for future population-level vascular analysis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SWITCH_014.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=E77u3qv9Fj
BibTex
@InProceedings{SobJan_Decoupled_MICCAISAT2026,
author = { Sobisch, Jannik AND Fuchs, Mirco AND Špiclin, Žiga AND Bizjak, Žiga},
title = { { Decoupled Node Localization and Edge Assignment for Vascular Graph Reconstruction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}