Tools for Automated Histology Image Analysis

In this thesis, we present three image processing tools inspired by and designed for histology image analysis. Histology, which is the examination of biological tissue under a microscope, is a critical technique in biomedical research and clinical practice. While slide preparation and imaging is increasingly becoming automated, the analysis of the resulting histology images is not: even routine analyses still require the trained eyes of a pathologist. In our work, we aim to understand histology images as a class of signals and develop tools to aid in the automated analysis of these signals. Our first contribution is in the area of histology image normalization, where the goal is to digitally remove the variation in staining between histology images, an important preprocessing step in many histology image analysis systems. To this end, we created a new benchmark dataset with which to compare normalization methods and proposed our own. Our second contribution is a tissue segmentation method, which delineates single-tissue regions in histology images. Along with this method, we propose a new mathematical model for histology images. Our final contribution is a method for describing distributions of angles, which is useful for segmentation as well as a variety of other image processing tasks.