Big data convergence of pathology and omics for disease prognosis

Case Western Reserve University, Cleveland OH

Investigators

Linked publications, trials & patents

Abstract

? DESCRIPTION (provided by applicant) There is a pressing need for improved means of interrogating cancer biology in order to identify markers associated with disease aggressiveness and patient outcome. While for a number of cancers, pathologic grade (morphologic appearance of cancerous tissue as assessed by a pathologist) has been found to be highly correlated with disease outcome, pathologic grade tends to suffer from significant inter-observer variability. Additionally pathologic grade alone is not useful in scenarios where two tumors may have subtle differences in their morphologic phenotype but significantly different behavior and outcome. Additionally while a number of molecular, gene expression based assays have been proposed for predicting outcome in cancers, the relatively poor to at best moderate success for these assays suggests that a solely omics driven approach for prognosis is not an optimal strategy. For most cancers, no single biomarker to date has been identified that is able to accurately and consistently stratify disease risk. This suggests a strong need for analytic and computational tools for quantitatively mining and integrating histologic image and molecular biomarkers to create fused predictors of disease risk and outcome. Such a unified approach is especially needed in cases where molecularly or morphologically similar tumors might have significantly different outcomes. This project will focus on the development of novel big data tools for processing of two key large scale data streams: 1) the high resolution (gigabyte-sized) digital images which capture pathology architecture and tissue morphology, and 2) a large set (up to tens of thousands) of molecular markers (e.g. NF-kB/p65/RelA, p-Akt (Ser473), periostin, cacna1d, ezh2, her2neu, ki67, propsa, and propsa2) found within the disease site. The ability to mine this information via innovative big data tools will allow for the creation of fused predictors of outcome and disease aggressiveness. A central hypothesis of this project is that the combination of quantitative histomorphometric and molecular features will yield a more predictive assay for evaluating disease aggressiveness compared to any single biomarker. This project will develop and evaluate the big data analysis and fusion tools in the context of evaluating disease aggressiveness for prostate cancer. Our over-arching goal is to translate big data tools to process and integrate imaging and molecular markers extracted from diseased tissue outlined by the following aims, Aim 1: Computer vision and machine learning tools for mining sub- visual image features associated with disease aggressiveness, Aim 2: Creating a fused predictor of disease aggressiveness by combining quantitative histomorphometric and molecular measurements, Aim 3: Evaluating the tools developed in Aims 1 and 2 for distinguishing between indolent and aggressive prostate cancer on data acquired from patients from across the leading urology institutions in the US including Johns Hopkins, The Cleveland Clinic, University Hospitals at CWRU, and University of Pennsylvania.