SDF-TopoNet: A Two-Stage Framework for Tubular Structure Segmentation via SDF Pre-training and Topology-Aware Fine-Tuning
Siyi Wu, Leyi Zhao, Haotian Ma, Xinyuan Song
TL;DR
SDF-TopoNet addresses the challenge of topologically faithful tubular structure segmentation by introducing a two-stage training regime that first learns a signed distance function (SDF) through MSE pre-training to encode topology and geometry, then fine-tunes with a dynamic adapter and a refined topology-aware loss. The dynamic adapter maps the learned SDF to binary segmentation, while topological losses based on persistence diagrams (Wasserstein or Betti matching) enforce global structural consistency with reduced computational burden. Empirical results on DRIVE, CREMI, Roads, and Elegans demonstrate improved Dice, IoU, and clDice scores over baselines, with ablations confirming the value of SDF pre-training for topological fidelity. The approach achieves superior topological accuracy without sacrificing pixel-level detail, offering a scalable framework for incorporating topology into segmentation models.
Abstract
Accurate segmentation of tubular and curvilinear structures, such as blood vessels, neurons, and road networks, is crucial in various applications. A key challenge is ensuring topological correctness while maintaining computational efficiency. Existing approaches often employ topological loss functions based on persistent homology, such as Betti error, to enforce structural consistency. However, these methods suffer from high computational costs and are insensitive to pixel-level accuracy, often requiring additional loss terms like Dice or MSE to compensate. To address these limitations, we propose \textbf{SDF-TopoNet}, an improved topology-aware segmentation framework that enhances both segmentation accuracy and training efficiency. Our approach introduces a novel two-stage training strategy. In the pre-training phase, we utilize the signed distance function (SDF) as an auxiliary learning target, allowing the model to encode topological information without directly relying on computationally expensive topological loss functions. In the fine-tuning phase, we incorporate a dynamic adapter alongside a refined topological loss to ensure topological correctness while mitigating overfitting and computational overhead. We evaluate our method on five benchmark datasets. Experimental results demonstrate that SDF-TopoNet outperforms existing methods in both topological accuracy and quantitative segmentation metrics, while significantly reducing training complexity.