Objective: A method for segmenting MR images of the breast was applied to determine fatty and fibroglandular tissue volumes in breasts of women in different age groups. The results were compared with subjective assessments of breast density from X-ray mammograms in the same patients.
Materials and methods: Two experienced mammographers assessed the percentage of fat in the breasts of 40 women who were 20-83 years old. MR images were obtained on a 1.0-T scanner equipped with a bilateral receive-only breast coil. Images were acquired using a three-dimensional T1-weighted gradient-echo sequence with a 1.25 x 1.4 x 2.5 mm resolution. On average, breast parenchyma appeared in 30 images in each breast. Image segmentation was based on a semiautomated, two-compartmental (fatty and fibroglandular tissue) model that accounts for partial volume effects. To validate the accuracy of the MR imaging segmentation technique, we performed a phantom study using an identical imaging sequence.
Results: The accuracy of the MR imaging segmentation of the phantom was of the order of 2%. In our subjects, fat content was 42.5% +/- 30.3% (mean +/- SD) on mammography versus 66.5% +/- 18% on MR images. Although we found a significant correlation (r = .63) between the two techniques, mammography poorly differentiated breasts containing less than 45% fat. When our analysis included only dense breasts (i.e., those containing less than 75% fat on MR images), the correlation coefficient decreased to .34. The largest discrepancies between mammography and MR imaging occurred in breasts that had 60-80% fat as measured on MR imaging.
Conclusion: Fatty and fibroglandular tissue can be differentiated on MR images of the breast with high precision and accuracy, therefore allowing assessment of breast density. The conclusions of researchers who used mammographic density patterns should be reassessed.