Histopathological image analysis with connections to genomics

The fields of imaging and genomics in cancer research have been mostly studied independently, but recently available datasets have made investigation into the synergy of these two fields possible. This work demonstrates the efficacy of computational histopathological image analysis to extract meaningful quantitative nuclear and cellular features from hematoxylin and eosin stained images that have meaningful connections to genomic data. Additionally, with the advent of whole slide images, significantly more data representing the variation in nuclear characteristics and tumor heterogeneity is available, which can aid in developing new analytical tools, such as the proposed convolutional neural network for nuclear segmentation, which produces state-of-the-art segmentation results on challenging cases seen in normal pathology. This robust segmentation tool is essential for capturing reliable features for computational pathology. Additionally, whole slide images capture rich spatial information about tumors, which presents a challenge, but also an opportunity for the development of new image processing tools to capture this spatial information, which could be considered for future work. Other histopathological image modalities and relevant machine learning tools are also considered for elucidating cellular processes of cancer