Segmentation and quantitative characterization of breast masses imaged using digital mammography

𝑬𝒏𝒈𝒍𝒊𝒔𝒉 Breast cancer is the leading cause of cancer death among women. Screening Mammography is the most effective method currently available for early detection of breast cancer. When breast cancer is detected at an early stage the prognosis is good because the tumour is smaller and more often well-differentiated, and less likely to have spread to regional lymph nodes. Computed radiography and direct digital detector mammography imaging systems provide a wide dynamic range for proper display of different densities of breast tissue areas. Their response over a wide range of X-ray intensities is linear; consequently, small differences between the attenuation coefficients of breast structures over a wide range of densities are clearly displayed. This includes the low signal areas associated with high densities found within tumours. Some masses infiltrate the surrounding breast tissue hence they exhibit ill- defined and intensity inhomogeneous boundaries with rough contour, while other masses exhibit well-defined edges and in most cases they possess smooth, round or oval shapes with macro-lobulations. The morphologic features of a mass such as its shape, margin and density give a clue to its benign or malignant nature. This study investigates and quantifies the changes in shape-based descriptors due to changes in the location of the initial level set contour in region based active contour models in delineating mammographic masses and proposes new methods to eliminate contour leakage and contour traps in active contour segmentation models which are due to intensity inhomogeneity within tumours and boundary regions of tumours. Furthermore, the study proposes a contextual region of interest model to assess the variation of texture features from the core to its periphery of biopsy proven malignant masses as a concept of tumour modelling in mammography and also the variation of texture features between grade 2 and grade 3 masses as a concept of tumour grading in mammography with texture analysis. ___________________________________________________________________