Quantitative Analysis of Digitized Histopathology Images with Applications to Prostate Cancer Screening
Advances in digital pathology technology have made possible the development of computer-aided diagnosis (CAD) systems based on quantitative analysis of high-content histopathology images. CAD systems have the potential to produce fast, objective, and consistent diagnosis outcomes to support pathologists in diagnosing several diseases in daily practice. The goal of this dissertation is the development of a framework for automated detection and grading of prostate cancer (PCa) regions from digitized histopathology based on the Gleason grading system. The main contributions of this work are fivefold. First, we present a comprehensive literature review of state-of-the-art systems and methods for quantitative analysis of digitized prostate histopathology using texture and morphology tissue descriptors. Second, we propose a fuzzy color standardization method that reduces the color variations observed in digitized histopathology images by locally transferring color statistics from a reference image to a target image without significantly altering the structure of the standardized image. Color standardization is a preprocessing step that allows us to analyze images from different sources. Third, we develop a system for automated classification of preselected cancerous regions belonging to Gleason grades 3, 4, and 5. This system assembles a set of binary classifiers trained on newly-developed tissue descriptors including color and wavelet-based features for robust cancer grading. Fourth, we propose a framework for automated detection and grading of cancerous regions from digitized tissue microarray cores and whole-slide images based on a combination of morphology and texture features. Finally, we statistically analyze the impact of using the proposed decision support tools on the percentage of agreement between pathologists in the assignment of Gleason grades to tissue samples.