Evaluation of diagnostic clues in histopathology through image processing techniques

Abstract

The primary method for the diagnostic interpretation of histopathologic sections is visual analysis. However, in a small, but significant percentage of cases, histopathologists do not come to a consensus. Therefore, due to the importance of early and accurate detection of tissue changes indicative of pathology, quantitative image analysis techniques have been applied to this problem. The accurate segmentation of image structures such as cells and glands in histopathological sections, as with all "natural scenes", proves challenging. This has led to the development of an additional segmentation technique, the heuristic gradient search. Following the successful segmentation and labeling of scene objects, algorithms evaluating diagnostic clues as to the shape, size and distribution of image components were developed in order to form an overall diagnosis. A description of these diagnostic clues and the image processing techniques residing in the computer vision system used to evaluate them are presented.