Comparison of breast density assessments based on interactive thresholding and automated fast fuzzy c-means clustering in three-dimensional MR imaging

摘要

The risk of breast cancer is increased by a number of factors including the breast density, considered to be the proportion of the fibroglandular tissue in the breast. Breast density can be assessed in three-dimensional breast MR images. This involves analysis of a large volume of image data. Most MR based density estimation methods are designed to work on images acquired using a particular MR scanner and field strength. Those that can be implemented on a wide range of scanner platforms require complex sequences; phantom scans or calibration which makes them impracticable for routine clinical use.In this study, two density estimation methods that can be applied to routine clinical practice, that have been developed in our institute, are compared. Prior to density estimation, the breast region anterior to the pectoral muscle on each breast is segmented separately using active contouring with gradient vector flow. The first method is based on interactive intensity thresholding applied to the pre-contrast T_1 weighted images corrected for coil sensitivity non-uniformities using proton density weighted images. The second method utilizes dual phase T_1 histogram based fuzzy maps computed from the proton density weighted and dynamic contrast enhanced T_1 weighted images. The fibroglandular tissues are detected from the maps by applying a fuzzy threshold.Analyses using multi centre data collected during the UK multi-centre study of MRI screening for breast cancer show high correlations between the breast densities estimated by both methods. The correlation is highest when a fuzzy threshold of 0.2 is used (r=0.943). For this value, estimated densities are almost 30% lower than those estimated by the first method. The second method is highly automated and increases the reproducibility and the consistency in estimations.