SDF-TopoNet: A Two-Stage Framework for Tubular Structure Segmentation via SDF Pre-training and Topology-Aware Fine-Tuning

2025-03-14

AI Summary

This work addresses the challenge of simultaneously achieving topological correctness and pixel-level accuracy in tubular structure segmentation (e.g., vessels, neurons), compounded by the high computational cost of persistent homology-based topological losses. We propose a two-stage paradigm: first, pretraining with signed distance fields (SDFs) to enhance geometric representation; second, topology-aware fine-tuning using dynamic LoRA adapters and a refined Betti number-based topological loss. Our approach avoids expensive persistent homology computations while preserving topological fidelity and improving pixel-level sensitivity. Evaluated on five benchmark datasets, the method achieves statistically significant gains in both topological accuracy and Dice coefficient over state-of-the-art methods. Training efficiency improves by 37%–52%, marking the first end-to-end tubular segmentation framework that is simultaneously efficient, accurate, and topologically reliable.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Learning with Manifolds
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

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 extbf{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.

Problem

• Accurate segmentation of tubular structures like blood vessels and neurons.
• Ensuring topological correctness with computational efficiency in segmentation.
• Reducing computational costs and improving pixel-level accuracy in segmentation.

Innovation

• Two-stage training strategy enhances segmentation accuracy
• SDF pre-training encodes topological information efficiently
• Dynamic adapter refines topological loss, reduces overfitting