Exploratory Identification of Image-Based Biomarkers for Solid Mass Pulmonary Tumors

If imaging is to serve as a valid biomarker in the assessment of the response of cancer to therapies, a reproducible and predictive radiologic metric is required. A biomarker is an indicator of a biological property that can be used to measure the progress of disease. While current size-based, quantitative techniques provide numerical representations of tumors, they are not necessarily indicative of disease progression for advanced cancers. In this paper, we present an end-to-end process to explore the use of other image-based features especially statistical textural features for cancer change detection. We exploit the earth mover's distance metric for measuring the change in the tumor burden over a period, between the time the baseline scans were taken, and the time the therapy response scans were taken. The time-to-progression (TTP) of the disease is our known patient outcome. We analyze the correlations between TTP and our change measurements and discover that the local texture energy feature is most predictive of disease progression, more so than the tumor burden size on which current quantitative measures are made.