Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Costal Cartilage Segmentation with Topology Guided Deformable Mamba: Method and Benchmark

Senmao Wang, Haifan Gong, Runmeng Cui, Boyao Wan, Yicheng Liu, Zhonglin Hu, Haiqing Yang, Jingyang Zhou, Bo Pan, Lin Lin, Haiyue Jiang

TL;DR

This work tackles costal cartilage segmentation from CT by introducing topology-guided deformable Mamba (TGDM), a two-stage framework that leverages centerline topology priors to model long-range voxel relationships with linear-time complexity. A lightweight U-Net provides coarse localization, followed by TGDM modules (PSM and GDM) that refine centerline coordinates and capture interactions across CC segments, aided by a bipartite matching loss for alignment. The authors also establish CCSeg, a comprehensive benchmark with 165 annotated cases and an out-of-domain test set, to rigorously evaluate segmentation methods. Empirical results show state-of-the-art Dice and NSD scores with high computational efficiency, demonstrating robustness across domains and offering a valuable resource for future CC segmentation research.

Abstract

Costal cartilage segmentation is crucial to various medical applications, necessitating precise and reliable techniques due to its complex anatomy and the importance of accurate diagnosis and surgical planning. We propose a novel deep learning-based approach called topology-guided deformable Mamba (TGDM) for costal cartilage segmentation. The TGDM is tailored to capture the intricate long-range costal cartilage relationships. Our method leverages a deformable model that integrates topological priors to enhance the adaptability and accuracy of the segmentation process. Furthermore, we developed a comprehensive benchmark that contains 165 cases for costal cartilage segmentation. This benchmark sets a new standard for evaluating costal cartilage segmentation techniques and provides a valuable resource for future research. Extensive experiments conducted on both in-domain benchmarks and out-of domain test sets demonstrate the superiority of our approach over existing methods, showing significant improvements in segmentation precision and robustness.

Costal Cartilage Segmentation with Topology Guided Deformable Mamba: Method and Benchmark

TL;DR

This work tackles costal cartilage segmentation from CT by introducing topology-guided deformable Mamba (TGDM), a two-stage framework that leverages centerline topology priors to model long-range voxel relationships with linear-time complexity. A lightweight U-Net provides coarse localization, followed by TGDM modules (PSM and GDM) that refine centerline coordinates and capture interactions across CC segments, aided by a bipartite matching loss for alignment. The authors also establish CCSeg, a comprehensive benchmark with 165 annotated cases and an out-of-domain test set, to rigorously evaluate segmentation methods. Empirical results show state-of-the-art Dice and NSD scores with high computational efficiency, demonstrating robustness across domains and offering a valuable resource for future CC segmentation research.

Abstract

Costal cartilage segmentation is crucial to various medical applications, necessitating precise and reliable techniques due to its complex anatomy and the importance of accurate diagnosis and surgical planning. We propose a novel deep learning-based approach called topology-guided deformable Mamba (TGDM) for costal cartilage segmentation. The TGDM is tailored to capture the intricate long-range costal cartilage relationships. Our method leverages a deformable model that integrates topological priors to enhance the adaptability and accuracy of the segmentation process. Furthermore, we developed a comprehensive benchmark that contains 165 cases for costal cartilage segmentation. This benchmark sets a new standard for evaluating costal cartilage segmentation techniques and provides a valuable resource for future research. Extensive experiments conducted on both in-domain benchmarks and out-of domain test sets demonstrate the superiority of our approach over existing methods, showing significant improvements in segmentation precision and robustness.
Paper Structure (21 sections, 13 equations, 4 figures, 3 tables)

This paper contains 21 sections, 13 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Costal Cartilage Segmentation
  4. Medical Volume Segmentation
  5. Prior Guided Segmentation
  6. Methods
  7. Overview
  8. Coarse Segmentation with Light-weight U-Net
  9. Topology-Guided Deformable Mamba
  10. Mamba: A Powerful State Space Model at Linear Complexity:
  11. PSM: Position Shifting Mamba for Coordinates Priors:
  12. GDM: Grouped Deformable Mamba for Long-range Relationship Modeling
  13. Overall loss function and network structure
  14. CCSeg: Benchmarking Costal Cartilage Segmentation
  15. Motivation
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Autologous applications and pathological changes of CC.
  • Figure 2: Overall pipeline of the proposed method. The stage 1 extracts the bounding box of the CC to help the network identify the approximate region of the CC. The stage 2 involves semantic segmentation of the CC based on topological structure prior knowledge. Using the results from the first stage and the corresponding CC location templates, we obtain the coordinates of each CC centerline through a mamba. Then, based on these coordinates, we use mamba for feature extraction to achieve more efficient interaction results of CC features.
  • Figure 3: The proposed grouped deformable mamba attention mechanism compared to previous sequential or self-attention mechanisms. Our deformable attention mechanism consists of two parts: coordinate acquisition based on position-prior and the grouped mamba attention mechanism.
  • Figure 4: Visual comparisons between our method and other state-of-the-art methods on CCSeg test set. Different categories of CC are denoted by different colors. The wrong prediction are marked by red boxes.