[1575] Validation of MMP-14 Quantitative Staining Evaluation in Ovarian Cancer Using Image Analysis

Patrice Desmeules, Dominique Trudel, Isabelle Bairati, Michele Orain, Bernard Tetu. University Hospital Center (CHU de Quebec), Quebec City, QC, Canada; University Health Network, Toronto, ON, Canada

Background: Visual immunostaining quantification is laborious and subject to significant intra and inter-observer variability. Automated image analysis is growing as an interesting and more objective tool for biomarker studies, especially when using tissue microarrays.
Design: Immunohistochemistry for the collagenase MMP-14 (Pierce, Rockford, Illinois) was performed on tissue microarrays built from 236 ovarian carcinomas. Cytoplasmic immunostaining was scored visually by two observers using a five-tiered scale (0%, 1-20%, 21-40%, 41-60%, 61-80%, 81-100%) and positive staining was graded into three intensity classes. Immunostaining evaluation was then submitted to automated image analysis on a random core sample using the CaloPix™ software (developed by TRIBVN, distributed by Agfa HealthCare). A protocol aimed at segmenting the tissue samples to exclude the non-epithelial component was first developed followed by an automated analysis of the percentage of positive epithelial cells and a three-tiered intensity scale.
Results: Although ovarian carcinomas have a high level of tissue heterogeneity, automated tissue segmentation was adequate in 49 out of 58 cores. Partial manual tissue segmentation was needed on the other cases. Differences in percentages of positive immunostained cell between manually and automated segmented zones varied between 1.2 and 17 % (median 3.2%). Overall, the percentage of automated stained cells ranged from 2.2 to 79.3%. We observed a linear correlation between results obtained by image analysis and visual scoring, despite threshold differences between both methods. With regard to intensity detection, a statistically significant correlation was also found between visual and automated scoring (Chi2: 12.4, p-value: 0,002).
Conclusions: Our study shows a good correlation between visual and automated image analysis evaluation for both staining proportion and intensity of MMP-14 immunostaining of the epithelial component despite the high complexity of ovarian carcinomas. Studies are underway to compare visual and automated scoring with survival data.
Category: Informatics

Monday, March 4, 2013 9:30 AM

Poster Session I Stowell-Orbison/Surgical Pathology/Autopsy Awards Poster Session # 247, Monday Morning


Close Window