Atlas-based automatic breast MRI segmentation using pectoral muscle and chest region model

Abstract

Accurate breast MRI segmentation is an important processing step in Computer Aided Diagnosis (CAD) systems and breast density assessment. Most of the atlas-based breast segmentation methods employ breast area as the template. Instead, we use both pectoral muscle and chest region model as the template, because there is great variability in breast shape and signal intensity. Pectoral muscle and chest region place in similar locations with similar shape and signal intensity. We demonstrate the high quality of the defined template for our atlas-based system. The presented approach is validated with a dataset of 2800 bilateral axial breast MR images from 50 women that include all of Breast Imaging Reporting and Data System (BI-RADS) breast density range. Five quantitative metrics as Dice Similarity Coefficient (DSC), Jaccard Coefficient (JC), total overlap, False Negative (FN) and False Positive (FP) are computed to compare similarity between automatic and manual segmentations. Our proposed algorithm obtains DSC, JC, total overlap, FN and FP values of 0.85, 0.75, 0.83, 0.16 and 0.11, respectively.