gms | German Medical Science

Introduction/background: Prostate cancer is routinely graded according to the Gleason grading scheme [1]. This scheme is predominantly based on textural appearance of the aberrant glandular structures. Gleason grading is difficult to standardize and leads to discussions due to notorious inter-rater and intra-rater disagreement. While a healthy glandular tissue has a regular shape that reflects its function, cancerous tissue has lost this regular structure. It was our objective to investigate whether digital image based histomorphometry quantifying theses structural changes could be used to achieve more standardized and reproducible classification outcomes.

Material and methods: In a proof-of-principle study we have developed a method to evaluate digitized histological images of single prostate cancer regions of interest (ROIs) in hematoxylin-eosin stained sections. Pre-processed color images were subjected to a segmentation algorithm using color deconvolution [2] providing binary (black & white) images. Resulting neoplastic epithelial gland-related objects were morphometrically quantified by two quantitative and objective geometric measures called inverse solidity and inverse compactness. We applied the procedure to prostate cancer probes of 125 patients. Each probe was independently classified for Gleason grade 3, 4 or 5 by an experienced pathologist blinded for the image analysis outcomes. For classification, multiple logistic regression models were used in order to find out whether and how well Gleason grades can be predicted from the logarithmized inverse compactness and inverse solidity values alone.

Results: A multivariate analysis reveals that both morphometric measures carry useful information for the construction of a morphometry-based Gleason grades classifier, so we found that logarithmized inverse compactness and inverse solidity together provide a powerful classifier distinguishing histologies of Gleason grade 3 from grades 4/5. We have obtained a correct classification of 94.4%. Furthermore, we showed the robustness of our classifier for variations in the parameter choice.

Discussion/conclusions: The results suggest that two simple and intuitively understandable measures obtained by image-based analysis and referring to both the size and the contour of segmented, gland-related objects seem to enable a good and reliable discrimination of low and high Gleason grades. Though the method still needs to be validated on an independent larger series of specimens, we believe that our analysis opens the way to a formalized, standardized digital morphometry of prostate cancer sections. In future work, instead of further restricting to ROIs, our method needs to be scaled up and extended to full morphometric scoring using information from several simultaneous core biopsies or complete histological slides.