Unencrypted login | home

Program Information

Breast Density Measurement with Cone-Beam CT and MRI: A Postmortem Study


H Ding

H Ding1*, T Johnson1, M Lin2, L Su2, S Molloi1, (1) Department of Radiological sciences, Unviersity of California, Irvine, CA, (2) Tu & Yuen Center for Functional Onco-Imaging, Radiation Oncology, Unviersity of California, Irvine, CA.

TH-A-103-11 Thursday 8:00AM - 9:55AM Room: 103

Purpose: To investigate the feasibility of breast density quantification using a cone-beam computed tomography (CBCT) and a magnetic resonance imaging (MRI) system with a postmortem study.

Methods: Twenty pairs of postmortem breasts were imaged with a bench-top CBCT system based on a CsI flat panel detector and an Aurora 1.5T dedicated Breast MRI system. The CT images were acquired at 50 kVp with a mean glandular dose (MGD) of approximately 3 mGy. A T1-weighted gradient echo technique was used in MRI scans to maximize the contrast between the adipose and glandular tissues. Image processing procedures were implemented to segment the glandular tissue from adipose tissue. For CBCT images, standard histogram thresholdings based on visual assessment and automated segmentation based on a basic fuzzy c-means (FCM) algorithm were used. For MRI images, FCM-based clustering and the coherent local intensity clustering (CLIC) method, which estimated and removed the bias field, were used. Finally, the breast densities measured with the two systems were compared to the definitive tissue compositions obtained from chemical analysis. The linear correlation coefficient, Pearson r, was used to evaluate the two imaging modalities and the investigated image segmentation methods.

Results: Both histogram thresholdings and FCM techniques used for CBCT images show excellent correlations with respect to the fibroglandular ratio (FGR) from chemical analysis, with r values of 0.964 and 0.980, respectively. The correlations of the MRI measurement were slightly worse than those from CBCT. The CLIC method successfully removed the intensity inhomogeneity induced by the bias field, and improved the r value from 0.864 to 0.915.

Conclusion: CBCT can be used to accurately quantify the breast density with automated clustering algorithms. It is anticipated that CBCT may outperform breast MRI in breast density quantification due to high spatial resolution and the absence of the bias field.

Contact Email: