Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BreastSegNet: Multi-label Segmentation of Breast MRI

Qihang Li, Jichen Yang, Yaqian Chen, Yuwen Chen, Hanxue Gu, Lars J. Grimm, Maciej A. Mazurowski

TL;DR

This work tackles the limited scope of breast MRI segmentation by introducing BreastSegNet, a nine-label, multi-label segmentation framework that encompasses tissues such as fibroglandular tissue, vessels, muscle, bone, lesion, lymph node, heart, liver, and implant. The authors curate a large annotated dataset via an iterative, model-assisted annotation pipeline and benchmark nine segmentation architectures across three categories, identifying nnU-Net with ResEncM as the top performer with an average Dice of 0.694 across all labels. The study provides detailed dataset and annotation procedures, implementation details, and open-source code, aiming to enable richer quantitative analyses of breast tissues and parameters from MRI. Despite strong performance on large structures, lymph nodes remain challenging due to rarity and appearance; the work lays groundwork for future improvements and data sharing to support broader breast MRI analysis and clinical applications.

Abstract

Breast MRI provides high-resolution imaging critical for breast cancer screening and preoperative staging. However, existing segmentation methods for breast MRI remain limited in scope, often focusing on only a few anatomical structures, such as fibroglandular tissue or tumors, and do not cover the full range of tissues seen in scans. This narrows their utility for quantitative analysis. In this study, we present BreastSegNet, a multi-label segmentation algorithm for breast MRI that covers nine anatomical labels: fibroglandular tissue (FGT), vessel, muscle, bone, lesion, lymph node, heart, liver, and implant. We manually annotated a large set of 1123 MRI slices capturing these structures with detailed review and correction from an expert radiologist. Additionally, we benchmark nine segmentation models, including U-Net, SwinUNet, UNet++, SAM, MedSAM, and nnU-Net with multiple ResNet-based encoders. Among them, nnU-Net ResEncM achieves the highest average Dice scores of 0.694 across all labels. It performs especially well on heart, liver, muscle, FGT, and bone, with Dice scores exceeding 0.73, and approaching 0.90 for heart and liver. All model code and weights are publicly available, and we plan to release the data at a later date.

BreastSegNet: Multi-label Segmentation of Breast MRI

TL;DR

This work tackles the limited scope of breast MRI segmentation by introducing BreastSegNet, a nine-label, multi-label segmentation framework that encompasses tissues such as fibroglandular tissue, vessels, muscle, bone, lesion, lymph node, heart, liver, and implant. The authors curate a large annotated dataset via an iterative, model-assisted annotation pipeline and benchmark nine segmentation architectures across three categories, identifying nnU-Net with ResEncM as the top performer with an average Dice of 0.694 across all labels. The study provides detailed dataset and annotation procedures, implementation details, and open-source code, aiming to enable richer quantitative analyses of breast tissues and parameters from MRI. Despite strong performance on large structures, lymph nodes remain challenging due to rarity and appearance; the work lays groundwork for future improvements and data sharing to support broader breast MRI analysis and clinical applications.

Abstract

Breast MRI provides high-resolution imaging critical for breast cancer screening and preoperative staging. However, existing segmentation methods for breast MRI remain limited in scope, often focusing on only a few anatomical structures, such as fibroglandular tissue or tumors, and do not cover the full range of tissues seen in scans. This narrows their utility for quantitative analysis. In this study, we present BreastSegNet, a multi-label segmentation algorithm for breast MRI that covers nine anatomical labels: fibroglandular tissue (FGT), vessel, muscle, bone, lesion, lymph node, heart, liver, and implant. We manually annotated a large set of 1123 MRI slices capturing these structures with detailed review and correction from an expert radiologist. Additionally, we benchmark nine segmentation models, including U-Net, SwinUNet, UNet++, SAM, MedSAM, and nnU-Net with multiple ResNet-based encoders. Among them, nnU-Net ResEncM achieves the highest average Dice scores of 0.694 across all labels. It performs especially well on heart, liver, muscle, FGT, and bone, with Dice scores exceeding 0.73, and approaching 0.90 for heart and liver. All model code and weights are publicly available, and we plan to release the data at a later date.

Paper Structure

This paper contains 16 sections, 1 equation, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Dataset
  3. Data Selection
  4. Annotation Pipeline
  5. Segmentation Details
  6. Methods
  7. Segmentation methods
  8. Evaluation metrics
  9. Experiments and Results
  10. Implementation Details
  11. Results
  12. Discussion
  13. Model Architecture Details
  14. U-Net
  15. Foundation model
  16. ...and 1 more sections

Figures (4)

  • Figure 1: The source data population, and the selected data distribution.
  • Figure 2: The data annotation pipeline.
  • Figure 3: Annotation visualization for nine anatomical labels: For each label, two representative slices are shown: the original image (rows 1 and 3) and its corresponding segmentation overlaid in orange (rows 2 and 4). For small or sparse structures (Vessel, Lesion, LN), zoomed-in views are provided in dashed red boxes, with the original regions marked by solid red boxes.
  • Figure 4: Qualitative comparison: Segmentation results of nnU-Net ResEncM, the best-performing model, compared to ground truth (GT) across nine anatomical labels. Rows show the original image, GT (green), and prediction (red). Zoomed-in views are shown for small structures (Vessel, Lesion, LN).