OBJECTIVE: To evaluate the ability of computer-assisted quantitative nuclear grading (QNG) using a microspectrophotometer and morphometry software to differentiate Feulgen-stained nuclei captured from normal urothelium, low grade transitional cell carcinoma (LG-TCC) and high grade transitional cell carcinoma (HG-TCC) cytology specimens. STUDY DESIGN: Feulgen-stained nuclei from a series of normal volunteers (urologic disease-free history) and from biopsy-confirmed cases of LG-TCC and HG-TCC were evaluated using a CAS-200 image analysis system. Thirty-eight nuclear morphometric descriptors (NMDs) were measured for each nucleus using a software conversion system. Backwards stepwise logistic regression analysis was applied to assess which of the NMDs contributed to QNG statistical models that could differentiate between nuclei from normals vs. LG-TCC, normals vs. HG-TCC, and LG-TCC vs. HG-TCC. Receiver operating characteristic curves and areas under the curve (AUC), as well as cell classification accuracy, were used to assess these differences. RESULTS: Statistically significant differences (P < .0001) were observed between all three categories. In the LG-TCC vs. normals, the QNG solution model required 16/38 features, with an AUC = 93%, a sensitivity = 85%, specificity = 86%, positive predictive value (PPV) = 87% and negative predictive value (NPV) = 84%. The QNG solution model for normals vs. HG-TCC required 12/38 nuclear features yielding an AUC = 99%, sensitivity = 99%, specificity = 98%, PPV = 98% and NPV = 99%. The QNG solution model for LG-TCC vs. HG-TCC required 17/38 nuclear features, with an AUC = 99%, sensitivity = 96%, specificity = 97%, PPV = 97% and NPV = 96%. CONCLUSION: Computer-assisted QNG cell classifiers based upon the measurement of 38 nuclear features, including size, shape and chromatin organization, are capable of differentiating normal urothelial nuclei from LG-TCC and HG-TCC nuclei as well as LG-TCC from HG-TCC nuclei. The QNG cell classifier has shown conclusively that there are morphometric differences between normal urothelial and LG-TCC nuclei that may not be apparent to the naked eye and that it may be useful in helping the pathologist determine the presence or absence of LG-TCC in bladder cytology specimens.
OBJECTIVE: To evaluate the ability of computer-assisted quantitative nuclear grading (QNG) using a microspectrophotometer and morphometry software to differentiate Feulgen-stained nuclei captured from normal urothelium, low grade transitional cell carcinoma (LG-TCC) and high grade transitional cell carcinoma (HG-TCC) cytology specimens. STUDY DESIGN: Feulgen-stained nuclei from a series of normal volunteers (urologic disease-free history) and from biopsy-confirmed cases of LG-TCC and HG-TCC were evaluated using a CAS-200 image analysis system. Thirty-eight nuclear morphometric descriptors (NMDs) were measured for each nucleus using a software conversion system. Backwards stepwise logistic regression analysis was applied to assess which of the NMDs contributed to QNG statistical models that could differentiate between nuclei from normals vs. LG-TCC, normals vs. HG-TCC, and LG-TCC vs. HG-TCC. Receiver operating characteristic curves and areas under the curve (AUC), as well as cell classification accuracy, were used to assess these differences. RESULTS: Statistically significant differences (P < .0001) were observed between all three categories. In the LG-TCC vs. normals, the QNG solution model required 16/38 features, with an AUC = 93%, a sensitivity = 85%, specificity = 86%, positive predictive value (PPV) = 87% and negative predictive value (NPV) = 84%. The QNG solution model for normals vs. HG-TCC required 12/38 nuclear features yielding an AUC = 99%, sensitivity = 99%, specificity = 98%, PPV = 98% and NPV = 99%. The QNG solution model for LG-TCC vs. HG-TCC required 17/38 nuclear features, with an AUC = 99%, sensitivity = 96%, specificity = 97%, PPV = 97% and NPV = 96%. CONCLUSION: Computer-assisted QNG cell classifiers based upon the measurement of 38 nuclear features, including size, shape and chromatin organization, are capable of differentiating normal urothelial nuclei from LG-TCC and HG-TCC nuclei as well as LG-TCC from HG-TCC nuclei. The QNG cell classifier has shown conclusively that there are morphometric differences between normal urothelial and LG-TCC nuclei that may not be apparent to the naked eye and that it may be useful in helping the pathologist determine the presence or absence of LG-TCC in bladder cytology specimens.