Public Domain Image Analysis Program Can Quantitate Nuclear Immunostains as Accurately as Proprietary Software
Dmitry A Turbin, Brenda Smith, Samuel Leung, Allen M Gown, Torsten O Nielsen, Diana N Ionescu. University of British Columbia, Vancouver, BC, Canada; Genetic Pathology Evaluation Centre, Vancouver, BC, Canada; BC Cancer Agency, Vancouver, BC, Canada; PhenoPath Laboratories, Seattle, WA
Background: Various nuclear immunostains including estrogen (ER) and progesterone (PR) receptor and Ki67 are routinely quantitated in the pathology laboratories as predictive or prognostic markers. A number of commercial slide scanners with proprietary image analysis software are available for automated quantitation, but are expensive and in most cases restricted to images produced by specific scanners.
Design: We created a public domain image analysis program entitled Subcellular Stain Analyzer (SCSA), a Java written plugin for ImageJ, which can be run on any Java-enabled operating system. We compared the accuracy and efficiency of automated quantitation of PR immunostain between SCSA and the Aperio IHC Nuclear Image Analysis algorithm (Aperio, Vista, CA) on triplicate core tissue microarrays (TMAs) built from 310 resection specimens of breast carcinomas clinically tested for ER, PR and HER2 at the British Columbia Cancer Agency recently. The TMAs slides were digitized using Aperio Scanscope; areas of invasive carcinoma were manually selected for each tissue core and at least 50 cells per core were analyzed. PR quantitation with the Aperio IHC Nuclear Image Analysis algorithm and SCSA provided continuous output of percent of positive cells, which was further ranked according to the principles of visual scoring. Statistical analysis was performed using PASW 17.0 (IBM Corporation, Armonk, NY).
Results: Automated unsupervised image analysis provided similar results between the two image analysis applications, with a Pearson correlation coefficient of 0.946 (p < 0.000001). When continuous scores were dichotomized into negative (0-1% positive cells) and positive (>1% positive cells), kappa statistics between two programs was 0.721.
Conclusions: SCSA, based on public domain image analysis software, may be as accurate and efficient in the quantification of PR immunostaining, when compared with the commercially available Aperio IHC Nuclear Image Analysis algorithm, and may offer a very cost effective alternative in both clinical and research settings. Future studies will involve comparison of quantitative image analysis of ER staining using these two applications with visual scores reported by pathologists.
Category: Quality Assurance
Monday, March 19, 2012 1:00 PM
Poster Session II # 274, Monday Afternoon