Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A2DMN: Anatomy-Aware Dilated Multiscale Network for Breast Ultrasound Semantic Segmentation

Kyle Lucke, Aleksandar Vakanski, Min Xian

TL;DR

A2DMN addresses the challenge of segmenting breast tissues in ultrasound images by introducing an anatomy-aware smoothness loss and a dilated multiscale network (DME blocks) built on an ESTAN backbone. By encoding tissue continuity and leveraging multiscale context, the method achieves finer tissue-boundary delineation and improved boundary metrics (HD/AAD) across mammary, muscle, tumor, and surrounding tissues. The approach uses binary-to-semantic transfer learning and five-class segmentation, reporting state-of-the-art performance on several tissue categories and robust boundary precision on a dataset of 325 BUS images. This work advances practical BUS analysis by providing anatomically informed segmentation with improved boundary reliability, potentially aiding breast density estimation and malignancy assessment.

Abstract

In recent years, convolutional neural networks for semantic segmentation of breast ultrasound (BUS) images have shown great success; however, two major challenges still exist. 1) Most current approaches inherently lack the ability to utilize tissue anatomy, resulting in misclassified image regions. 2) They struggle to produce accurate boundaries due to the repeated down-sampling operations. To address these issues, we propose a novel breast anatomy-aware network for capturing fine image details and a new smoothness term that encodes breast anatomy. It incorporates context information across multiple spatial scales to generate more accurate semantic boundaries. Extensive experiments are conducted to compare the proposed method and eight state-of-the-art approaches using a BUS dataset with 325 images. The results demonstrate the proposed method significantly improves the segmentation of the muscle, mammary, and tumor classes and produces more accurate fine details of tissue boundaries.

A2DMN: Anatomy-Aware Dilated Multiscale Network for Breast Ultrasound Semantic Segmentation

TL;DR

A2DMN addresses the challenge of segmenting breast tissues in ultrasound images by introducing an anatomy-aware smoothness loss and a dilated multiscale network (DME blocks) built on an ESTAN backbone. By encoding tissue continuity and leveraging multiscale context, the method achieves finer tissue-boundary delineation and improved boundary metrics (HD/AAD) across mammary, muscle, tumor, and surrounding tissues. The approach uses binary-to-semantic transfer learning and five-class segmentation, reporting state-of-the-art performance on several tissue categories and robust boundary precision on a dataset of 325 BUS images. This work advances practical BUS analysis by providing anatomically informed segmentation with improved boundary reliability, potentially aiding breast density estimation and malignancy assessment.

Abstract

In recent years, convolutional neural networks for semantic segmentation of breast ultrasound (BUS) images have shown great success; however, two major challenges still exist. 1) Most current approaches inherently lack the ability to utilize tissue anatomy, resulting in misclassified image regions. 2) They struggle to produce accurate boundaries due to the repeated down-sampling operations. To address these issues, we propose a novel breast anatomy-aware network for capturing fine image details and a new smoothness term that encodes breast anatomy. It incorporates context information across multiple spatial scales to generate more accurate semantic boundaries. Extensive experiments are conducted to compare the proposed method and eight state-of-the-art approaches using a BUS dataset with 325 images. The results demonstrate the proposed method significantly improves the segmentation of the muscle, mammary, and tumor classes and produces more accurate fine details of tissue boundaries.
Paper Structure (15 sections, 1 equation, 2 figures, 4 tables)

This paper contains 15 sections, 1 equation, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Dataset
  5. Network Architecture
  6. Smoothness Loss
  7. Transfer Knowledge From Binary Segmentation
  8. Results
  9. Training Settings and Evaluation Metrics
  10. The Effectiveness of the DME Blocks
  11. The Effectiveness of the Smoothness Loss
  12. Overall performance
  13. Conclusion
  14. Compliance with Ethical Standards
  15. Acknowledgments

Figures (2)

  • Figure 1: A2DMN architecture.$\oplus$ is the concatenation operator. $A$ denotes kernel size, $C$ defines the number of kernels, $D$ denotes the dilation rate. DME blocks represent a Dilated Multiscale ESTAN block.
  • Figure 2: A2DMN with and without the proposed smoothness loss. The violet dashed boxes represent misclassified regions.