55 Development of Voxelized Patient Models for Investigation of Image Quality and Absorbed Dose within the Glandular Breast Tissue on Chest CT Scans

Faculty Mentor: Frederic Noo (Radiology & Imaging Sciences, University of Utah)

Recent publications have shown that breast cancer can be incidentally detected on chest CT scans ordered for other purposes. However, little is known about dose absorption and image quality in the breasts from these exams. Furthermore, for spatial resolution reasons, the breasts (or, more specifically, the glandular breast tissue) are often not included in the reconstructed field of view. Our goal is to understand these aspects through development of voxelized patient models and ultimately their use in virtual clinical trials. Here, we report on this development. Voxelized patient models, created from patient CT data, are computational models of a body region divided into small volumes or voxels, where each voxel corresponds to a specific tissue type.

Four voxelized patient variability, we selected four patients that cover important variations in breast size, breast density, and body habitus models were developed from patient CT data gathered from The Cancer Imaging Archive. For this purpose, and to be able to account for anatomical. Using threshold values suggested in the literature, all pixels within the CT chest images were sorted into indices 1-16, where each index number corresponds to a unique tissue type. Index 0 was retained for air. Next, masks covering the breast area in each CT slice needed to be created. To do this, the CT data was processed in MATLAB™. Then, all breast pixels were sorted into two additional indices, index 17 representing fat and index 18 representing glandular breast tissue. To avoid artificial material characterization due to noise, a median filter was applied to the breasts to refine the segmentation between fat and glandular breast tissue. The x-ray mass attenuation coefficients of all atoms found in human tissues were collected from NIST, and this data was interpolated over energy values from 10 keV to 150 keV by steps of 1 keV. The chemical compositions of the tissues and their nominal mass density were taken from the literature and were used with the mass attenuation coefficients to produce the x-ray linear attenuation coefficients of the tissues over the same energy values. Then, from the linear attenuation coefficients, virtual CT images of the models at 78.5 keV were generated and compared with the original CT images. Visual inspection of the images was done as well as quantitative assessment using profiles. Results are very encouraging. The new phantoms are now deemed ready to investigate image reconstruction aspects like noise and view sampling as well as to investigate scatter and absorbed radiation dose with Monte-Carlo transport.

BIBLIOGRAPHY