Breast tissue description and modeling in mammography

This dissertation presents an approach to modeling, implementation, and evaluation of simulated mammography and synthetic digital mammograms. Breast tissue distribution, mammographic compression, and x-ray image acquisition are modeled separately. The 3-D breast tissue model captures large and medium scale properties of the breast, including regions of predominantly adipose and fibroglandular tissue, adipose compartments, and ductal network. Synthetic mammograms are computed by applying the image acquisition model to the synthetically compressed breast model.; The computer implementation of the breast model consists of a regular 3-D array of voxels with elastic and x-ray attenuation properties of the corresponding tissues. Simulation parameters can be adjusted to vary the distribution of the model elements, allowing for anatomic variations in normal tissue. The compression model is implemented as a separate analysis of breast sections. The image acquisition model was adopted from the literature, assuming a monoenergetic x-ray approximation and a parallel beam geometry without scatter.; Clinically acquired galactograms were analyzed to evaluate the ductal model by comparing real and synthetic duct branching patterns. The compression model was tested by comparing the real and synthetic force needed for the same compressed breast thickness. Since no clinically available 3-D breast imaging techniques match the resolution of mammography, the tissue model was evaluated by comparing synthetic acid real mammogram texture using morphological size analysis, texture energy, and fractal analysis. Synthetic mammograms with varying adipose compartments sizes were generated and compared with real mammograms from the MIAS database at 200 μm/pixel resolution. The best matching was achieved for the simulated adipose compartments with radii of 4–13.3 mm (2.7–5.33 mm) for predominantly adipose (fibroglandular) tissue.; We have been able to develop a 3-D breast model based on the understanding of the macroscopic tissue organization and generate synthetic mammograms using a mammography simulation. By varying the distribution of simulated structures we have been able to match the values of several texture features averaged over a large number of mammograms. Differences between the synthetic and real images are more evident for the features emphasizing smaller spatial scales. We expect an improved local similarity with the introduction of more detailed tissue structures in the breast model.