Validation of automated image analysis methods for evaluation of AMACR expression in prostate biopsies with high grade prostatic intraepithelial neoplasia (HGPIN)

A49

Introduction
 >Automated image analysis (AIA) systems for quantification of immunostained tissue could provide more accurate and sensitive results than manual scoring by a pathologist. Previously we have shown that AMACR expression in normal glands from patients subsequently diagnosed with prostate cancer was higher than those who did not. No difference was observed for AMACR expression in HGPIN glands. In this study, we hypothesize that the difference in AMACR expression between the two subsets of patients might be better quantified using AIA.
 >Materials and Methods
 >Prostatic needle biopsies from 44 patients with HGPIN were immunostained for AMACR. 21 patients (cases) showed cancer on a later biopsy and 22 (controls) had no cancer and at least 3 negative biopsies. Immunostained slides were digitally scanned using the ScanScope CS ® (Aperio Technologies, CA). Regions of interests were drawn within benign and HGPIN compartments and scored using ‘Positive Pixel Count™’ and ‘Co-Localization’ algorithms. Separate scores for benign and HGPIN glands were computed as the product of percent area positive and mean intensity of staining. Manual scoring was performed on an ordinal scale of 0-3 intensity by a single pathologist.
 >Results
 >For the normal compartment, the mean manual score per gland for the cases (0.29) was higher than that of the controls (0.21) [p<0.05]. AIA score for the normal compartment was higher in the cases than controls, but not significant (p=0.07). For the HGPIN compartment, mean manual score was 1.25 and 1.18 for the cases and controls respectively [p= 0.77]. However, AIA score within the HGPIN compartment of cases (12.52) was significantly higher than controls (5.62) [p=0.01].
 >Conclusions
 >The ability to discriminate HGPIN according to risk of subsequent cancer supports the hypothesis that AIA has enhanced sensitivity and biological validity. This method might provide improved measurements for research and clinical prediction.