OBJECTIVES: Despite the introduction of the new World Health Organization 2004 grading classification, the grading of urothelial carcinoma remains difficult and subjective. The aim of this study was to evaluate the role of computer-assisted image morphometric analysis as a tool to improve the objectivity of histologic grading of urothelial carcinoma. METHODS: A total of 75 urinary bladder biopsies from a cohort of patients with a first-time diagnosis of urothelial carcinoma representing low-grade (n = 19) and high-grade (n = 56) urothelial carcinoma were evaluated. Quantitative nuclear morphometry was performed on these biopsies using approximately 80 to 100 cells per case. A total of 17 nuclear morphometry features were extracted, and a bootstrap-based predictor selection using stepwise logistic regression analysis was performed. Subsequently, a validation was performed using the five top features from the logistic regression analyses by implementing a nonparametric discriminant analysis to identify the most discriminative features that predicted for high-grade cases. RESULTS: The bootstrap technique included nuclear pleomorphism as the most frequently selected predictor of high-grade urothelial carcinoma (in 213 of 500 replicates). Validation using the top five features in the logistic regression analysis method (pleomorphism, configuration run length, DNA mass, feret-Y, and age) using discriminant analysis gave a resubstitution error of 4%, indicating the usefulness of the selected predictors. CONCLUSIONS: The present study is the first to provide a morphometric validation of the World Health Organization 2004 system for pathologic grading of bladder cancer. Furthermore, quantitative nuclear morphometry could aid in the objective grading of urinary bladder biopsies. This information might aid the treating physicians in better risk stratification of patients with urothelial carcinoma.
OBJECTIVES: Despite the introduction of the new World Health Organization 2004 grading classification, the grading of urothelial carcinoma remains difficult and subjective. The aim of this study was to evaluate the role of computer-assisted image morphometric analysis as a tool to improve the objectivity of histologic grading of urothelial carcinoma. METHODS: A total of 75 urinary bladder biopsies from a cohort of patients with a first-time diagnosis of urothelial carcinoma representing low-grade (n = 19) and high-grade (n = 56) urothelial carcinoma were evaluated. Quantitative nuclear morphometry was performed on these biopsies using approximately 80 to 100 cells per case. A total of 17 nuclear morphometry features were extracted, and a bootstrap-based predictor selection using stepwise logistic regression analysis was performed. Subsequently, a validation was performed using the five top features from the logistic regression analyses by implementing a nonparametric discriminant analysis to identify the most discriminative features that predicted for high-grade cases. RESULTS: The bootstrap technique included nuclear pleomorphism as the most frequently selected predictor of high-grade urothelial carcinoma (in 213 of 500 replicates). Validation using the top five features in the logistic regression analysis method (pleomorphism, configuration run length, DNA mass, feret-Y, and age) using discriminant analysis gave a resubstitution error of 4%, indicating the usefulness of the selected predictors. CONCLUSIONS: The present study is the first to provide a morphometric validation of the World Health Organization 2004 system for pathologic grading of bladder cancer. Furthermore, quantitative nuclear morphometry could aid in the objective grading of urinary bladder biopsies. This information might aid the treating physicians in better risk stratification of patients with urothelial carcinoma.