Quantitative nuclear morphometry, Markovian texture descriptors, and DNA content captured on a CAS-200 Image analysis system, combined with PCNA and HER-2/neu immunohistochemistry for prediction of prostate cancer progression.

One hundred and twenty-four localized prostate cancer patients operated on at Johns Hopkins Hospital (JHH) since 1975 were identified. The sample was optimized for evaluation of prostate cancer progression. Based upon accurate clinical histories, these radical prostatectomy patients included 50 progressors and 74 non-progressors using appearance of serum PSA as an indication of recurrence (mean follow-up = 8.6 +/- 1.8 years, range 7-15 years). All patients included in the study had no involvement of their seminal vesicles or lymph nodes at the time of prostatectomy. Average time to progression was 3.6 +/- 2 years, range of 1-8 years. Using paraffin-embedded specimens, several five micron sections were cut and placed on Probe-On slides; one slide was H&E-stained and the other was Feulgen-stained. The H&E and Feulgen-stained slides were screened and "dotted" by pathologists at JHH and CytoDynostics, Inc. A CAS-200 Image analysis system (Cell Image Systems, Elmhurst, IL) equipped with a Cell Measurement Program version 1.2 beta, was used to capture the Feulgen-stained images and to perform the calculations. From the "dotted" areas, 150 cancer cells were selected for measurement of DNA content and 27 nuclear morphometric shape and size factors, including 21 Markovian chromatin texture variables. Additional sections were used for immunochemistry staining with an alkaline phosphatase streptavidin-biotin complex stain to detect and quantitate cancer cells binding monoclonal antibodies directed against proliferating cell nuclear antigen (PCNA) and HER-2/neu antigen. All data were entered into a statistical program (STATA) for further analysis and univariate and multivariate statistical analysis was performed using logistic regression and its stepwise variant. The biomarkers of greatest utility to detect progressors when analyzed univariately included post-operative Gleason score (p = < 0.0001), HER-2/neu antigenicity (p = 0.0147), CAS-200 DNA ploidy (p = 0.008), and twelve Markovian nuclear texture and shape features (p = < 0.0001), whereas PCNA (p = 0.160) failed. The optimal set of nuclear morphometry progression tumor features were selected using backward stepwise logistic regression estimate analysis which drops variables due to collinearity. Although post-operative Gleason score is a strong univariate predictor of progression, DNA ploidy and HER-2/neu contributed significantly to further stratification of higher risk groups within the low Gleason score subpopulation. The best Markovian features combined with post-operative Gleason score generated sensitivity = 90%, specificity = 96%, positive predictive value = 94%, negative predictive value = 93% and the area under the receiver operator curve was 0.975.