Towards a quantitative grading of bladder tumors.

The histological inspection of tumor tissue for the purpose of reporting a tumor grade is a problem of significant clinical importance. The grading by a pathologist is only partly reproducible due to vaguely defined, subjective criteria. In this article we describe and evaluate a set of measurable features that quantitate the differences in tumor tissue. Different aspects of the reproducibility of the measurements under varying conditions of image selection, focus, and noise have been investigated. Three hundred thirty-three images were digitized from 111 bladder tissue sections (4 microns thick, Feulgen stained), using the ICAS microscope-camera platform. A segmentation routine was developed to segment the images into nuclei and background without any user interaction. Size, shape, optical density, and texture features were measured on and among the objects found by this segmentation routine using the image analysis package Acuity. The results of the measurements showed that there is a significant quantitative difference between grade 1 and grade 3 tumors. Grade 2 tumors can be described as "in between grade 1 and grade 3" and falling somewhere on an increasing scale between grades 1 and 3. Grade 2 tumors do not seem to represent a statistically distinct population. The procedure described here is shown to be quite reproducible in the presence of noise, reasonably reproducible in the event of a modest amount of defocusing (with grade 3 tumors exhibiting the most sensitivity), and less reproducible in the context of which fields-of-view are chosen for analysis.