LAUSR

1864 Computer Extracted Features of Nuclear Shape and Architecture Predict Oncotype DX Risk Categories for Early Stage ER+ Breast Cancer Jon Whitney, Andrew Janowczyk, German Corredor, Hannah Gilmore, Anant Madabhushi. Case Western, Cleveland, OH. Background: Oncotype DX (ODX) is a 21 gene assay used to stratify women with early stage ER+ breast cancer into low, intermediate and high risk categories, high risk being women who would benefi t from adjuvant chemotherapy. In this work we explored the use of computer extracted features of nuclear architecture and morphology from routine H&E images to predict ODX risk categories for lymph node negative ER+ BCa. Design: The dataset contains 178 BCa patients with H&E stained whole slide images of low, intermediate, and high ODX risk. Nuclei were segmented and classifi ed as either epithelial or stromal using Deep Learning models, and had nuclear architectural features (median, standard deviation of the edge lengths of nuclear graphs), as well as nuclear shape features extracted.