An automated framework for brain vessel centerline extraction from CTA images

TL;DR

The paper tackles brain vessel centerline extraction from CTA by introducing a four-component framework that uses lumen segmentation generated from annotated centerlines to supervise a dual-branch topology-aware UNet (DTUNet) trained with a topology-aware loss (TAL). During training, both annotated centerlines and generated lumen segmentation are leveraged to enforce topological consistency, while inference proceeds without lumen generation, reducing overhead. Extensive multi-center evaluation shows improved average symmetric centerline distance (ASCD) and overlap (OV) over state-of-the-art methods, with favorable subgroup results suggesting clinical utility for stroke assessment. The approach, including the public code, offers a scalable path toward reliable, topology-preserving vessel centerlines from CTA for cerebrovascular analysis.

Abstract

Accurate automated extraction of brain vessel centerlines from CTA images plays an important role in diagnosis and therapy of cerebrovascular diseases, such as stroke. However, this task remains challenging due to the complex cerebrovascular structure, the varying imaging quality, and vessel pathology effects. In this paper, we consider automatic lumen segmentation generation without additional annotation effort by physicians and more effective use of the generated lumen segmentation for improved centerline extraction performance. We propose an automated framework for brain vessel centerline extraction from CTA images. The framework consists of four major components: (1) pre-processing approaches that register CTA images with a CT atlas and divide these images into input patches, (2) lumen segmentation generation from annotated vessel centerlines using graph cuts and robust kernel regression, (3) a dual-branch topology-aware UNet (DTUNet) that can effectively utilize the annotated vessel centerlines and the generated lumen segmentation through a topology-aware loss (TAL) and its dual-branch design, and (4) post-processing approaches that skeletonize the predicted lumen segmentation. Extensive experiments on a multi-center dataset demonstrate that the proposed framework outperforms state-of-the-art methods in terms of average symmetric centerline distance (ASCD) and overlap (OV). Subgroup analyses further suggest that the proposed framework holds promise in clinical applications for stroke treatment. Code is publicly available at https://github.com/Liusj-gh/DTUNet.

Figures (10)

• Figure 1: An overview of the proposed framework. During the training phase, it consists of four major components, including pre-processing approaches, lumen segmentation generation, DTUNet, and post-processing approaches. Note that the proposed framework does not require the lumen segmentation generation during the inference phase.
• Figure 2: An illustration of the pre-processing approaches. (a) A raw CTA image; (b) an image after registration and masking; (c) a clipped and normalized image; (d) split patches.
• Figure 3: An illustration of the lumen segmentation generation. (a) An annotation of cerebral vessel centerlines where each centerline segment is assigned a different color; (b) a cross-sectional slice obtained along a centerline; (c) lumen segmentation obtained from a slice using graph cuts and robust kernel regression; (d) complete brain lumen segmentation created using an interpolation approach.
• Figure 4: A diagram of the dual-branch topology-aware UNet (DTUNet). Note that the predicted lumen segmentation patches $S'$ output by the segmentation branch are used to generate final vessel centerlines via a merging operation and post-processing approaches.
• Figure 5: A diagram of the fusion block. It is used to effectively aggregate the feature $F_{s}$ from the segmentation branch and the feature $F_{c}$ from the centerline branch.