Image Processing and Computer Aided Diagnosis in Computed Tomography of the Breast

Abstract : A novel breast cancer imaging technique -- dedicated cone-beam breast computerized tomography (CT) -- is currently under development. It is designed to deliver low-dose radiation to a patient while removing the superposition of breast tissues, which is a limiting factor of the conventional mammography technique. The new technique will particularly benefit women with dense breasts. The development of the breast CT imaging technique requires effective and efficient ways to reduce its scattered radiation as well as denoising so as to provide high-quality images to help radiologists make diagnostic decisions. Therefore, it is important to investigate different possible image processing tools for use with CT and decide which one is best based on image quality metrics such as contrast-to-noise ratio (CNR) and the observer performance study via a receiver operating characteristic (ROC) analysis. The raw breast CT data was successfully reconstructed via the Feldkamp filtered back projection (FBP) algorithm for cone-beam geometry. A Gaussian noise model taking into account the energy-integrating characteristic of a flat-panel detector was developed. Based on this model, the maximum likelihood estimate of the scatter-free image via the expectation maximization (EM) algorithm was derived. A partial diffusion equation-based image denoising technique also was implemented.