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

Conventional subjective visual grading of malignancy in bladder carcinoma suffers from poor standardization. In an attempt to improve this by image analysis we carried out a texture analysis of a total of 197 images from Feulgen-stained histological tissue sections. The textural features used were based on the spatial gray-tone co-occurrence probability matrices for step-lengths 3, 4 and 6 pixels. These features were compared to the subjective grading by multivariate statistical methods. Using a two-stage hierarchical classifier 84.3% agreement was obtained between the subjective and computer-based classification. This was verified through the jackknifed classification technique.