A publicly available deep learning model and dataset for segmentation of breast, fibroglandular tissue, and vessels in breast MRI

Christopher O. Lew,Majid Harouni,Ella R. Kirksey,Elianne J. Kang,Haoyu Dong,Hanxue Gu,Lars J. Grimm,Ruth Walsh,Dorothy A. Lowell,Maciej A. Mazurowski

DOI: https://doi.org/10.1038/s41598-024-54048-2

IF: 4.6

2024-03-07

Scientific Reports

Abstract:Breast density, or the amount of fibroglandular tissue (FGT) relative to the overall breast volume, increases the risk of developing breast cancer. Although previous studies have utilized deep learning to assess breast density, the limited public availability of data and quantitative tools hinders the development of better assessment tools. Our objective was to (1) create and share a large dataset of pixel-wise annotations according to well-defined criteria, and (2) develop, evaluate, and share an automated segmentation method for breast, FGT, and blood vessels using convolutional neural networks. We used the Duke Breast Cancer MRI dataset to randomly select 100 MRI studies and manually annotated the breast, FGT, and blood vessels for each study. Model performance was evaluated using the Dice similarity coefficient (DSC). The model achieved DSC values of 0.92 for breast, 0.86 for FGT, and 0.65 for blood vessels on the test set. The correlation between our model's predicted breast density and the manually generated masks was 0.95. The correlation between the predicted breast density and qualitative radiologist assessment was 0.75. Our automated models can accurately segment breast, FGT, and blood vessels using pre-contrast breast MRI data. The data and the models were made publicly available.

multidisciplinary sciences

What problem does this paper attempt to address?

The main goal of this paper is to address several key issues in breast density assessment:

1. **Insufficient Data Sharing**: In existing studies, the datasets used for training and evaluating algorithms are often not publicly available, which limits the development and comparison of algorithm models.

2. **Limitations of Quantification Tools**: Current methods for assessing breast density (whether qualitative or semi-automated) still have issues with inter-user consistency, and these methods are not precise enough in defining breast density, leading to limited comparability and accuracy of assessment results.

3. **Impact of Vascular Tissue**: In breast magnetic resonance imaging (MRI), vascular tissue may be mistaken for fibroglandular tissue (FGT), which can affect the accurate assessment of breast density.

To address these issues, the research team has undertaken the following work:

- Created and publicly released a large dataset that includes pixel-level annotations of the breast, fibroglandular tissue, and vessels, with clear annotation standards.

- Developed, evaluated, and publicly released an automated segmentation method based on convolutional neural networks for the segmentation of the breast, fibroglandular tissue, and vessels.

- This method achieved high Dice Similarity Coefficients (DSC) on the test set, reaching 0.92 for the breast, 0.86 for fibroglandular tissue, and 0.65 for vessels.

- The study also found that the correlation coefficient between the model-predicted breast density and manually generated masks was 0.95, and the correlation coefficient with radiologists' qualitative assessments was 0.75.

In summary, this study aims to improve the assessment of fibroglandular tissue on breast MRI by developing a publicly available deep learning model and its dataset, to achieve a more objective and efficient method for breast density assessment.