Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dynamic Position Transformation and Boundary Refinement Network for Left Atrial Segmentation

Fangqiang Xu, Wenxuan Tu, Fan Feng, Malitha Gunawardhana, Jiayuan Yang, Yun Gu, Jichao Zhao

TL;DR

This work tackles left atrial segmentation when inputs are randomly cropped, which disrupts object positioning and boundary continuity. It introduces DPBNet, a one-stage framework that combines a Shuffle-then-Reorder Attention Module (SRAM) to capture dynamic spatial dependencies and a Dual Fine-grained Boundary (DFB) loss to sharpen dual foreground/background boundaries. The backbone is a VNet, trained with a combined loss and evaluated on the 2018 LA dataset, where DPBNet achieves state-of-the-art performance and demonstrates robustness to random cropping. Ablation studies confirm the effectiveness of SRAM in improving target localization and the DFB loss in enhancing boundary precision, enabling practical, high-precision LA segmentation without heavy pre/post-processing.

Abstract

Left atrial (LA) segmentation is a crucial technique for irregular heartbeat (i.e., atrial fibrillation) diagnosis. Most current methods for LA segmentation strictly assume that the input data is acquired using object-oriented center cropping, while this assumption may not always hold in practice due to the high cost of manual object annotation. Random cropping is a straightforward data pre-processing approach. However, it 1) introduces significant irregularities and incompleteness in the input data and 2) disrupts the coherence and continuity of object boundary regions. To tackle these issues, we propose a novel Dynamic Position transformation and Boundary refinement Network (DPBNet). The core idea is to dynamically adjust the relative position of irregular targets to construct their contextual relationships and prioritize difficult boundary pixels to enhance foreground-background distinction. Specifically, we design a shuffle-then-reorder attention module to adjust the position of disrupted objects in the latent space using dynamic generation ratios, such that the vital dependencies among these random cropping targets could be well captured and preserved. Moreover, to improve the accuracy of boundary localization, we introduce a dual fine-grained boundary loss with scenario-adaptive weights to handle the ambiguity of the dual boundary at a fine-grained level, promoting the clarity and continuity of the obtained results. Extensive experimental results on benchmark dataset have demonstrated that DPBNet consistently outperforms existing state-of-the-art methods.

Dynamic Position Transformation and Boundary Refinement Network for Left Atrial Segmentation

TL;DR

This work tackles left atrial segmentation when inputs are randomly cropped, which disrupts object positioning and boundary continuity. It introduces DPBNet, a one-stage framework that combines a Shuffle-then-Reorder Attention Module (SRAM) to capture dynamic spatial dependencies and a Dual Fine-grained Boundary (DFB) loss to sharpen dual foreground/background boundaries. The backbone is a VNet, trained with a combined loss and evaluated on the 2018 LA dataset, where DPBNet achieves state-of-the-art performance and demonstrates robustness to random cropping. Ablation studies confirm the effectiveness of SRAM in improving target localization and the DFB loss in enhancing boundary precision, enabling practical, high-precision LA segmentation without heavy pre/post-processing.

Abstract

Left atrial (LA) segmentation is a crucial technique for irregular heartbeat (i.e., atrial fibrillation) diagnosis. Most current methods for LA segmentation strictly assume that the input data is acquired using object-oriented center cropping, while this assumption may not always hold in practice due to the high cost of manual object annotation. Random cropping is a straightforward data pre-processing approach. However, it 1) introduces significant irregularities and incompleteness in the input data and 2) disrupts the coherence and continuity of object boundary regions. To tackle these issues, we propose a novel Dynamic Position transformation and Boundary refinement Network (DPBNet). The core idea is to dynamically adjust the relative position of irregular targets to construct their contextual relationships and prioritize difficult boundary pixels to enhance foreground-background distinction. Specifically, we design a shuffle-then-reorder attention module to adjust the position of disrupted objects in the latent space using dynamic generation ratios, such that the vital dependencies among these random cropping targets could be well captured and preserved. Moreover, to improve the accuracy of boundary localization, we introduce a dual fine-grained boundary loss with scenario-adaptive weights to handle the ambiguity of the dual boundary at a fine-grained level, promoting the clarity and continuity of the obtained results. Extensive experimental results on benchmark dataset have demonstrated that DPBNet consistently outperforms existing state-of-the-art methods.
Paper Structure (14 sections, 8 equations, 3 figures, 2 tables)

This paper contains 14 sections, 8 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Shuffle-then-Reorder Attention module
  4. Shuffle-then-Reorder Operation.
  5. Shuffle-then-Reorder Attention Module.
  6. Dual Fine-grained Boundary Loss
  7. Experiments and results
  8. Dataset.
  9. Implementation Details.
  10. Comparison Results
  11. Ablation Study
  12. Conclusion
  13. Acknowledgments.
  14. Disclosure of Interests.

Figures (3)

  • Figure 1: Input comparison showcases center cropping in (a), (b), random cropping in (c), (d), with all cropping applied to the original image (e). Red areas represent ground truth, while orange arrows highlight irregular object and discontinuous boundary in random cropping outcomes.
  • Figure 2: The pipeline of our Dynamic Position transformation and Boundary refinement framework utilizes VNet as its backbone architecture. Our learning loss consists of a cross-entropy loss and a dual fine-grained boundary loss on the DFB map.
  • Figure 3: Visualization results by various methods and DPBNet, showcasing prediction (blue) and ground truth (red). First row shows 2D prediction-ground truth overlap, and second features 3D distance maps from prediction-ground truth differences.