Women with mammographically dense breasts have an increased risk for breast cancer.¹ Quantification of breast density with mammography has inherent limitations.^2,3 MRI-based methods to measure the amount and concentration of fibroglandular tissue (FGT) have potential advantages, including the fundamental ability to distinguish fibroglandular from adipose tissue, however variability has also been described between different MRI-based methods.^3-7 In this work we propose to use chemical-shift encoded (CSE)-MRI techniques^8-10 that correct for all known confounders of the MRI signal that could negatively impact the accuracy and precision of MR-based quantification of FGT. Thus, the purpose of this study is to demonstrate the feasibility of confounder-corrected CSE-MRI techniques to estimate the proton density water fraction (PDWF) in order to quantify the volume and concentration of FGT in the breast.

Six volunteers were consented to undergo confounder-corrected CSE-MRI at 3T (Discovery 750w, GE Healthcare, Waukesha, WI) of the bilateral breasts using an 8-channel phased array breast coil (GE Healthcare). Images were obtained with an investigational version of a commercially available sequence (IDEAL IQ, GE Healthcare) with a 32 cm axial field-of-view, bandwidth of ± 90.91, TR = 8.8ms, and six echoes (TEs = 1.0, 2.0, 3.1, 4.1, 5.2, and 6.2 ms). A low flip angle of 1^o was used to minimize T1 related bias (see Figure 1). Further, images were obtained with two protocols, each with different spatial resolution. Protocol 1: voxel 1.4 x 2.0 x 3.0 mm³ and Protocol 2: voxel 2.0 x 2.2 x 5.0 mm³. Scan time for Protocols 1 and 2 was 105s and 60s, respectively.

The process used to calculate the FGT concentration is illustrated in Figure 2. Separation of fat and water from the acquired echo images uses a complex fitting algorithm including correction for R2* (=1/T2*) and spectral modeling of fat.^11,12 The resulting fat and water separated images are used to generate the PDWF map, which we define as $$$ PDWF = \rho_{W} / (\rho_{W} + \rho_{F}) $$$, where $$$ \rho_{W} $$$ and $$$ \rho_{F} $$$ are the proton densities of water and fat, respectively. PDWF was calculated using magnitude discrimination to avoid noise bias.¹³ For this proof of concept implementation, the chest wall was excluded from the map through manual segmentation while skin removal was accomplished using an erosion operation in Matlab (MathWorks, Natick, MA). Segmentation was performed over a 6cm axial slab covering the central portion of the bilateral breasts for both protocols, producing the regions of interest (ROI)s used in this analysis. The total volume of both breasts (V_B) was calculated using the ROI. The volume of FGT (V_FGT) was calculated by integration of the PDWF maps over the ROI. Concentration of FGT is defined as V_FGT/V_B x 100. Bland-Altman analysis was performed to compare V_FGT between protocols.

Figure 3 lists the V_FGT and FGT concentrations for the six volunteers by protocol. Both volume and concentration of FGT correspond with expected results given the visual appearance of the FGT. Bland-Altman plot of the V_FGT is shown in Figure 4 demonstrating excellent agreement between methods.

The measured volume and concentration of FGT demonstrated excellent agreement between the two CSE-MRI protocols. Small differences are likely due to limitations in our manual segmentation technique, which resulted in inconsistent ROIs between studies and inadvertent inclusion of pectoralis muscle in some cases. This can be corrected by implementing an automated segmentation technique.

Our CSE-MRI method used to quantify the volume and concentration of FGT is based on a fundamental property of tissue, ie: PDWF. This approach is differs from the subjective evaluation of breast density in routine clinical practice, ie: the mammographic appearance of radiodense FGT relative to radiolucent fat, which is limited due to intra- and inter-reader variability¹⁴ and lack of quantitation. Quantitative mammographic methods are limited by 2D technique, breast compression/orientation, ionizing radiation, and differing acquisition parameters.^2,3 In the proposed MRI method, the PDWF reflects a quantitative measurement of FGT corrected for relevant confounding factors over a wide range of acquisition protocols. This technique eliminates the dependence on voxel size avoiding measurement error caused by partial volume averaging of fat and FGT (unavoidable in the breast, where fat and FGT are intermingled), and is thus robust to changes in spatial resolution, as demonstrated.

We have demonstrated the use of CSE-MRI to create a PDWF map and quantify the volume and concentration of FGT in the breast. Further studies are needed to confirm these results, improve the segmentation method, and determine if volume or concentration of FGT is the most predictive quantitative biomarker for risk of breast cancer.