Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GAPNet: Granularity Attention Network with Anatomy-Prior-Constraint for Carotid Artery Segmentation

Lin Zhang, Chenggang Lu, Xin-yang Shi, Caifeng Shan, Jiong Zhang, Da Chen, Laurent D. Cohen

TL;DR

The paper tackles automated carotid artery segmentation in MR BB-VWI, a task hindered by complex neck anatomy and atherosclerotic lesions. It introduces GAPNet, a two-stage granularity attention network guided by an anatomical prior derived from the isoperimetric theorem, and adds a topology-based penalty to enforce plausible vessel-wall shapes. The topology penalty uses the isoperimetric ratio $\mu(\mathfrak S)=\frac{4\pi\mathcal{A}(\mathfrak S)}{\mathcal{L}(\partial\mathfrak S)^2}$ with a threshold $\tau$, implemented as $\mathcal{P}_{\rm topology}=\max\{0,-(\mu(\mathfrak S)-\tau)\}$ to discourage leaks. Experiments on COSMOS 2022 and CAVWSC 2021 show GAPNet achieving state-of-the-art vessel-wall and lumen segmentation, evidencing improved accuracy and robustness for CA delineation in challenging neck anatomy.

Abstract

Atherosclerosis is a chronic, progressive disease that primarily affects the arterial walls. It is one of the major causes of cardiovascular disease. Magnetic Resonance (MR) black-blood vessel wall imaging (BB-VWI) offers crucial insights into vascular disease diagnosis by clearly visualizing vascular structures. However, the complex anatomy of the neck poses challenges in distinguishing the carotid artery (CA) from surrounding structures, especially with changes like atherosclerosis. In order to address these issues, we propose GAPNet, which is a consisting of a novel geometric prior deduced from.

GAPNet: Granularity Attention Network with Anatomy-Prior-Constraint for Carotid Artery Segmentation

TL;DR

The paper tackles automated carotid artery segmentation in MR BB-VWI, a task hindered by complex neck anatomy and atherosclerotic lesions. It introduces GAPNet, a two-stage granularity attention network guided by an anatomical prior derived from the isoperimetric theorem, and adds a topology-based penalty to enforce plausible vessel-wall shapes. The topology penalty uses the isoperimetric ratio with a threshold , implemented as to discourage leaks. Experiments on COSMOS 2022 and CAVWSC 2021 show GAPNet achieving state-of-the-art vessel-wall and lumen segmentation, evidencing improved accuracy and robustness for CA delineation in challenging neck anatomy.

Abstract

Atherosclerosis is a chronic, progressive disease that primarily affects the arterial walls. It is one of the major causes of cardiovascular disease. Magnetic Resonance (MR) black-blood vessel wall imaging (BB-VWI) offers crucial insights into vascular disease diagnosis by clearly visualizing vascular structures. However, the complex anatomy of the neck poses challenges in distinguishing the carotid artery (CA) from surrounding structures, especially with changes like atherosclerosis. In order to address these issues, we propose GAPNet, which is a consisting of a novel geometric prior deduced from.
Paper Structure (10 sections, 5 equations, 5 figures, 2 tables)

This paper contains 10 sections, 5 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Granularity Attention Network
  4. Penalty Terms from the Anatomical Prior
  5. Experimental Results
  6. Dataset
  7. Implementation Details
  8. Comparison with State-Of-The-Art Models
  9. Ablation Study
  10. Conclusion and Future Work

Figures (5)

  • Figure 1.1: Typical challenges in CA segmentation task from MRI images. The green arrows point to the vessel walls undergoing complex deformations due to lesions.
  • Figure 2.1: Diagram of the proposed GAPNet.The backbone of the network consists of two U-shaped networks embedded with the FRA module and the MIE module.
  • Figure 2.2: a and b respectively illustrate the normal and abnormal segmentation results CA. The regions indicated by cyan color are the segmented CA walls whose external boundary contours are indicated by black solid lines.
  • Figure 3.1: Comparison of different methods. Arrows indicate excessive segmentation and boxes denote incomplete segmentation structures or insufficient details.
  • Figure 3.2: Visualization of the heat map from the final layer of the decoder.