Segmentation of primary breast tumor nuclei in histological images

Breast cancer (BCa) is a heterogeneous and diverse disease. They are sub-classified into four major subtypes – luminal A, luminal B, Her2-overexpressing and basal-like. It has been seen that the phenotypic variability between BCa occurs within, but not across these major subtypes. These subtypes not only have distinct behaviors but also differ in responses to therapy which highlights the importance of identifying each subtype of BCa for appropriate therapeutic decisions. In order to determine pathologic staging, pathologists routinely evaluate various features like Regional lymph node metastasis status and histologic grade. Such histological analysis though useful and cost-effective, largely depends on the experience of the pathologist performing the analysis. To achieve a better reproducibility and reduced dependence on the pathologist, there is a need to develop a system to objectively predict tumor subtype which was previously possible only through expensive molecular testing and immunohistochemistry (IHC). In order to establish an Image analysis paradigm to generate predictions of tumor sub-type objectively, a reliable method to segment nuclei to analyze their properties individually has been discussed. The performance of this method is evaluated and is seen to perform better - qualitatively and quantitatively compared to a preliminary segmentation