Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RICAU-Net: Residual-block Inspired Coordinate Attention U-Net for Segmentation of Small and Sparse Calcium Lesions in Cardiac CT

Doyoung Park, Jinsoo Kim, Qi Chang, Shuang Leng, Liang Zhong, Lohendran Baskaran

TL;DR

The paper tackles lesion-specific CAC segmentation in non-contrast cardiac CT, where CAC lesions are small and sparsely distributed, causing severe class imbalance. It introduces RICAU-Net, a U-shaped network that embeds Coordinate Attention in both its encoder and decoder and leverages a weighted Focal LogDice loss to address imbalance and small lesion sizes. Empirical results show RICAU-Net achieving the highest per-lesion Dice scores across all four CAC classes compared with five U-Net baselines, with notable gains for the rare LM lesion. This approach enables more accurate lesion-specific CAC scoring, potentially enhancing CHD risk stratification and guiding targeted clinical interventions.

Abstract

The Agatston score, which is the sum of the calcification in the four main coronary arteries, has been widely used in the diagnosis of coronary artery disease (CAD). However, many studies have emphasized the importance of the vessel-specific Agatston score, as calcification in a specific vessel is significantly correlated with the occurrence of coronary heart disease (CHD). In this paper, we propose the Residual-block Inspired Coordinate Attention U-Net (RICAU-Net), which incorporates coordinate attention in two distinct manners and a customized combo loss function for lesion-specific coronary artery calcium (CAC) segmentation. This approach aims to tackle the high class-imbalance issue associated with small and sparse CAC lesions. Experimental results and the ablation study demonstrate that the proposed method outperforms the five other U-Net based methods used in medical applications, by achieving the highest per-lesion Dice scores across all four lesions.

RICAU-Net: Residual-block Inspired Coordinate Attention U-Net for Segmentation of Small and Sparse Calcium Lesions in Cardiac CT

TL;DR

The paper tackles lesion-specific CAC segmentation in non-contrast cardiac CT, where CAC lesions are small and sparsely distributed, causing severe class imbalance. It introduces RICAU-Net, a U-shaped network that embeds Coordinate Attention in both its encoder and decoder and leverages a weighted Focal LogDice loss to address imbalance and small lesion sizes. Empirical results show RICAU-Net achieving the highest per-lesion Dice scores across all four CAC classes compared with five U-Net baselines, with notable gains for the rare LM lesion. This approach enables more accurate lesion-specific CAC scoring, potentially enhancing CHD risk stratification and guiding targeted clinical interventions.

Abstract

The Agatston score, which is the sum of the calcification in the four main coronary arteries, has been widely used in the diagnosis of coronary artery disease (CAD). However, many studies have emphasized the importance of the vessel-specific Agatston score, as calcification in a specific vessel is significantly correlated with the occurrence of coronary heart disease (CHD). In this paper, we propose the Residual-block Inspired Coordinate Attention U-Net (RICAU-Net), which incorporates coordinate attention in two distinct manners and a customized combo loss function for lesion-specific coronary artery calcium (CAC) segmentation. This approach aims to tackle the high class-imbalance issue associated with small and sparse CAC lesions. Experimental results and the ablation study demonstrate that the proposed method outperforms the five other U-Net based methods used in medical applications, by achieving the highest per-lesion Dice scores across all four lesions.
Paper Structure (17 sections, 3 equations, 2 figures, 2 tables)

This paper contains 17 sections, 3 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Coordinate Attention Module
  4. Residual-block Inspired Coordinate Attention U-Net
  5. Class-imbalance Problem and Loss Function
  6. Class-imbalance Problem
  7. Customized Combo Loss Function
  8. Experiments and Results
  9. Dataset
  10. Preprocessing
  11. Implementation Details
  12. Evaluation Metric
  13. Comparison with Other Methods
  14. Ablation Study
  15. Conclusion
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Overview of the RICAU-Net architecture (left), and the CA module and the RICA block in RICAU-Net (right). The CA module illustrates the overall process of aggregating positional information from both horizontal and vertical directions of a feature map into channel attention. $r$ represents the reduction ratio, which is employed to decrease the complexity of the model by reducing the number of channels num27. In our architecture, we utilized $r$ = 32.
  • Figure 2: Visualization of the segmentation performance of RICAU-Net, U-Net, and Nested U-Net on three input images. The zoomed-in lesion images and corresponding per-lesion Dice scores with highlights are shown in the images. Our model significantly improves the segmentation performance for sparse and smaller CAC lesions, while also enhancing the results of other major ones.