Quantification of Volumetric Breast Density Using MRI data: Comparison between 3 Representative Slice Method and Whole Breast Density Measurement

Abstract

Introduction: Volumetric breast density measured from breast MRI is more suitable for density analysis than mammographic density. However, measurement of volumetric percentage density (VPD) from MRI of the whole breast is time consuming and impractical to apply for large breast MRI databases. The purpose of this study was to determine whether VPDs measured from 3 representative slices of MRI correlates with the VPDs measured from the whole breast. Methods: A total of 151 breast cancer patients who underwent bilateral breast MRI were included. Pre-contrast, T1 weighted, fat suppressed, sagittal images were used for VPD measurement. VPD of the contralateral normal breast was measured using MIPAV software (Version 5.3.3, NIH, USA). The borders of the breasts were outlined manually and fibroglandular tissue was segmented from the fat tissue using Fuzzy C-Means algorithm. The number of slices ranged from 30 to 72 on measuring VPD from the whole breast. The 3 representative slices selected for density measurements were the middle (n), (n+15)th and (n-15)th slices. VPD was measured from the whole breast as well as from 3 representative slices for each patient and the results compared using t-test, intra-class correlation coefficient (ICC) and coefficient of variation (CV). The average time taken to measure each breast by each method was calculated. Results: Mean VPD measured from 3 representative slices was not different from that measured from the whole breast (24.32% vs. 24.55%, p = 0.39

mean absolute difference = 2.72). The CV was 9.74%. VPDs measured by the two methods showed excellent agreement (ICC = 0.96

95% CI = 0.95- 0.97). Average time taken to measure VPD by 3-slice method was significantly lower than that by the whole breast measurement (133 sec vs. 572 sec

p < .0001). The mean absolute difference in VPD between the two methods was not significantly different for fatty (<20% VPD) and dense (>40% VPD) breasts (2.37% vs. 2.62%

p = 0.67). CV was higher for fatty breasts (15.33%) than dense breasts (5.11%). Conclusions: The 3 slice MRI VPD is in excellent agreement with the VPD measured from the whole breast MRI and is less time consuming