An Application Of Pattern Recognition To Epithelial Tissues

Subjective histologic criteria for the differential grading of tumors of the epitheliums of the human urinary bladder have been given quantitative explication in the interactive digital image processing and decision making computer programming system, PEEP/DECIDE/GRAPH. These criteria concern size, shape, stain content, orientation and texture of digital images of nuclei in histologic sections and have been shown to be consistent with cytology and survival. Two tissue sections differing in visual appearance were stained with hematoxylin to render nuclear detail with high contrast, and were scanned at 570 nm, 630x, and 0.5u resolution with a vibrating mirrow digital scanning microphotometer, leading to 1024 x 1024 pixel images. The two tissues were characterized by nuclear absorbance, nuclear area, nuclear density, density variance, density entropy, chromatin texture and distribution of angular nuclear orientation. Prior to this feature extraction and image characterization, algorithms for visual scene analysis were utilized to delineate individual cell nuclei. The extracted quantitative features were used in both supervised and unsupervised computer classification algorithms to determine whether the two tissue sections had consistent morphometric differences and warranted differential grading.