Breast imaging has improved the early detection of breast cancer thereby decreasing the mortality rate; however, thousands of women are wrongly diagnosed each year. Improving the sensitivity and specificity of breast cancer imaging is an important area of research and development. One of the major hurdles in imaging research arises from the difficulty in accruing human subject data because of cost, time, or patient risk considerations. Consequently, computerized phantoms are an important research tool that can help in developing new imaging techniques and devices. They can simulate a potentially unlimited amount of patient anatomies and provide a known truth with which to quantitatively evaluate, compare, and improve new imaging technologies in a cost-effective and efficient method. It is essential for computerized phantoms to be anatomically realistic and produce realistic imaging data such that results from studies utilizing the phantoms are indicative of what would occur in human subjects.;The purpose of this dissertation is to develop a three-dimensional computer generated breast phantom that is based on empirical data that can be used in breast imaging research. Currently available breast phantoms are either voxelized phantoms with fixed anatomy or flexible mathematical phantoms based on geometric primitives such as spheres and cylinders. In this work, we present the method to generate a suite of hybrid breast phantoms that combine the realism of a voxelized phantom with the flexibility to easily model anatomical variations by incorporating a mathematical basis.;The first step in phantom generation was to acquire and process the imaging data. We received dedicated breast computed tomography imaging data of pendant uncompressed breasts of human subjects from our collaborators at UC Davis. We implemented pre- and post-reconstruction algorithms to reduce the noise and scatter inherent in the images from the low-dose acquisition of the data and the cone-beam geometry of the CT system, respectively. Following image processing, we developed a custom volumetric segmentation algorithm to differentiate the breast tissues and maintain the high-resolution detail available in the imaging data. Derived from real human data, this step produced an anatomically realistic basis for the breast phantom.;Following segmentation, a subdivision surface model of the breast tissue was generated. This step introduced flexibility to the empirically based phantom by using a mathematical description for the breast tissue surfaces as subdivision surface models can be altered using affine or other transformations. This phantom can be used for imaging studies using an uncompressed geometry or it can be used to generate a finite element mesh of the breast to be used for a compression model.;Simulated compression of the breast phantom was achieved by applying finite element methods that can realistically deform the phantom. The material properties of the different types of breast tissue were incorporated into this model. Also, a comprehensive analysis of the different parameters that can affect breast phantom compression was performed. After simulated compression, the calculated deformations can be applied to the subdivision surface model to be used for studies on modalities with a compressed breast imaging geometry.;Simulated images can be generated directly from the subdivision surface model of the breast phantom with existing image simulation tools. We implemented an analytical projection algorithm that realistically models the x-ray imaging process and includes effects from quantum noise. Images were generated that exemplify the effect of different mechanical parameter assignments to the breast phantom tissue.;To further expand our database of imaging phantoms, we implemented deformation and morphing techniques to generate new and unique datasets from the limited number of original human subject datasets. In order to illustrate the full capabilities of the phantom, we generated simulated mammograms from several finite element compressed breast phantoms and are in the process of performing a user study to validate their level of realism. While early results from the phantom are promising, there are many future improvements to be made and futures studies to be conducted.;The presented work engenders a substantial advancement in tools for breast imaging research. We have successfully developed a suite of hybrid breast imaging phantoms that combine realistic anatomy with the flexibility of a mathematical approach. These phantoms can be used effectively for imaging research to develop and improve new imaging techniques and devices for breast cancer detection and prevention.
展开▼
