Breast Dense Tissue Segmentation

Breast Dense Tissue Segmentation segments dense tissue and the overall breast area in digital mammograms using a confusion matrix (CM)-YNet deep learning model to quantify breast density for breast cancer risk assessment.

Key Features:

• Automated segmentation: Automatically identifies and segments the overall breast area and dense tissue within digital mammograms.
• CM-YNet deep learning: Implements a confusion matrix (CM)-YNet model to handle noisy labels by modeling each radiologist's labels for improved ground-truth estimation.
• Breast detection and pectoral muscle exclusion: Includes explicit breast detection and exclusion of the pectoral muscle to restrict segmentation to relevant breast tissue.

Scientific Applications:

• Breast cancer risk assessment: Provides quantitative breast density measurements to support identification of individuals at elevated risk for breast cancer.
• Research and clinical studies: Produces reliable segmentation outputs for multi-center studies and clinical research on the relationship between breast density and cancer incidence.

Methodology:

The workflow uses a confusion matrix (CM)-YNet model that models each radiologist’s labels to address label variability, includes breast detection and pectoral muscle exclusion, and was validated on a multi-center mammogram corpus with an average DICE score of 0.82.