Grading of transitional cell bladder carcinoma by image analysis of histological sections

Image analysis of histological sections was used to achieve a more objective malignancy grading of transitional cell carcinoma of the bladder. Images from Feulgen-stained sections from a clinical material of 197 tumours were analyzed. Features at various levels of analysis, e.g. mainly related to individual objects, neighbouring objects and the entire image, were analyzed. The features used were based on relational rather than individual nuclear features. With this technique, typical tissue architecture and degree of order/disorder can be described. These characteristics were compared by means of multivariate statistical methods with the subjective grading of the pathologists at our institution. This comparison provided the best subset of features and the agreement between the subjective and computer-based classification was 73%. The size, orientation and variation of the grey scale of the nuclei were particularly powerful. On a continuous scale, from grade 1 to 2A and 2B to 3, the four grades formed two distinct classes, low and high grade. In this distribution, the intermediate grade 2A was identified as a displaced grade 1 and the other intermediate grade 2B as a displaced grade 3.