Computer-aided prognosis of ER+ breast cancer histopathology and correlating survival outcome with Oncotype DX assay

Abstract

The current gold standard for predicting disease survival and outcome for lymph node-negative, estrogen receptor-positive breast cancer (LN-, ER+ BC) patients is via the gene-expression based assay, Oncotype DX. In this paper, we present a novel computer-aided prognosis (CAP) scheme that employs quantitatively derived image information to predict patient outcome analogous to the Oncotype DX Recurrence Score (RS), with high RS implying poor outcome and vice versa. While digital pathology has made tissue specimens amenable to computer-aided diagnosis (CAD) for disease detection, our CAP scheme is the first of its kind for predicting disease outcome and patient survival. Since cancer grade is known to be correlated to disease outcome, low grade implying good outcome and vice versa, our CAP scheme captures quantitative image features that are reflective of BC grade. Our scheme involves first semi-automatically detecting BC nuclei via an Expectation Maximization driven algorithm. Using the nuclear centroids, two graphs (Delaunay Triangulation and Minimum Spanning Tree) are constructed and a total of 12 features are extracted from each image. A non-linear dimensionality reduction scheme, Graph Embedding, projects the image-derived features into a low-dimensional space, and a Support Vector Machine classifies the BC images in the reduced dimensional space. On a cohort of 37 samples, and for 100 trials of 3-fold randomized cross-validation, the SVM yielded a mean accuracy of 84.15% in distinguishing samples with low and high RS and 84.12% in distinguishing low and high grade BC. The projection of the high-dimensional image feature data to a 1D line for all BC samples via GE shows a clear separation between, low, intermediate, and high BC grades, which in turn shows high correlation with low, medium, and high RS. The results suggest that our image-based CAP scheme might provide a cheaper alternative to Oncotype DX in predicting BC outcome.