|NSECT | GSECT | Quant CT/Tomo | Dual Energy | Chest Tomo | Breast Tomo | Breast Density|
|Quant. Image | Emerg. Quant. Imaging | Perf. Metrology | Clinical Trials | Emerg. Clinical
Multi-modality breast density analysis
Many previous studies have established quantitatively that mammographic density is well correlated with cancer risk. The advantages of mammography include reasonable sensitivity, high resolution, and low patient dose. However, a mammogram still suffers from being a two-dimensional (2D) projection of the three-dimensional (3D) breast.
The goal of this project is twofold: (1) To develop multiple 3D imaging techniques for measuring breast density, and (2) to correlate those imaging markers with cytological ground truth for risk. This multidisciplinary study bridges the gap between imaging and biology, as well as between diagnosis and treatment. The project is a collaboration between Joseph Lo PhD (PI), Victoria Seewaldt MD (head of the Duke University Breast Cancer Prevention Clinic), Ehsan Samei PhD (director of RAI Labs), and Jay Baker MD (RAI Labs faculty member and head of the Duke Breast Imaging Clinic).
Breast MRI is often considered to be the imaging "gold standard" for determining breast density, because it is 3D and provides good contrast differentiation between fat vs.
fibroglandular tissue. We acquired MRI scans from a large cohort of 118 high-risk screening subjects enrolled in Dr. Seewaldt's on-going studies. We developed a fully-automated, "slice by slice" breast segmentation and classification paradigm, as illustrated in Figure 1. Once the breasts were successfully separated from the image background, an iterative, bimodal segmentation technique was employed to estimate the threshold that best separated fibroglandular vs fat voxel values. Figure 2 shows the resulting segmented MRI slice with fibroglandular voxels highlighted. Once the aforementioned methods were carried out for each slice, the total number of fibroglandular voxels throughout the entire volume divided by the breast volume yielded the volumetric breast density.
MRI has several key limitations, however, including high cost, limited availability, use of intravenous contrast, slow scan times, and difficulties in breast segmentation. We have therefore also investigated breast tomosynthesis as an alternative 3D imaging technique. Although considerable information is lost due to its narrow angular acquisition, breast tomosynthesis can still create a 3D volume at a dose comparable to that of traditional mammography. Our pilot study so far demonstrates encouraging correlation between density measurements made by tomo vs MRI, as shown in Figure 3.
In on-going work, we study the correlation between these multi-modality imaging techniques for measuring breast density. We will also compare these imaging markers with cytological ground truth from a procedure known as Random Periareolar Fine Needle Aspiration (RPFNA). Dr. Seewaldt uses this research procedure to serially sample breast cytology from high-risk women and to monitor response to prevention treatments such as Tamoxifen. Atypia is considered to be a major risk factor. Indeed, the presence of atypia in RPFNA has been prospectively validated to predict a 5.6-fold risk increase, and RPFNA has been validated as an accurate assessment of short-term breast cancer risk and response to chemoprevention in high-risk patients. Our long term goal is to develop inexpensive, noninvasive, fast, and reliable imaging markers for breast cancer risk.