Renal papillary carcinoma classification into subtypes may be reproduced by nuclear morphometry.

OBJECTIVE: To analyze relationships between nuclear features of papillary renal cell carcinoma (PapRCC) subtypes. STUDY DESIGN: The material for the study consisted of 53 cases, of which 29 were type 1, 17 type 2 and 7 intermediate. At least 100 nuclei per case were segmented from images of 4',6-diamidino-2-phenylindole-stained slides. The geometric and texture features were extracted and used for analysis. RESULTS: In analysis of variance, it was shown that both individual cases and tumor types differ in the majority of the parameters. On nonsupervised expectation-maximization clustering, it was possible to classify the nuclei into separate categories, but PapRCC classes were not reproduced. The neural network classified the nuclei with sensitivity >0.6 and specificity >0.75. Analyzing the results for individual cases, the nuclei of type 1 cases were properly classified in 74-91%, nuclei of type 2 cases in 58-80% and nuclei of intermediate cases in 53-70%. CONCLUSION: Our findings show that PapRCC subtypes are distinct enough to be reproduced by image analysis.