Please select Print from the file menu or right mouse click and choose print from the menu to print this page.

[1401] Automation of TMA Interpretation through a Combination of Epithelial-Recognition and Specific-Recognition Algorithms

SH Barsky, LM Tatke, SH Patil, RE Jimenez, S Mahooti, AS Basu, AS Gholap. The Ohio State University College of Medicine, Columbus, OH; BioImagene, Inc, Cupertino, CA

Background: Tissue microarrays (TMAs) are a form of high throughput screening akin to cDNA microarray and proteomic analyses, yet, unlike the latter types, are based on manual construction and subjective interpretation. Because of the increasing demand for TMAs predicted to occur over the next decade, we felt it necessary to investigate whether their interpretation could be completely automated.
Design: In this study we used TMAs made from breast, colon and lung cancer and analyzed each TMA for two nuclear, cytoplasmic and membrane immunocytochemical markers that were either homogeneously or heterogeneously expressed and compared the algorithmic measurements with the subjective ones of 3 pathologists. For the breast carcinoma TMA, the targets were: nuclear (ER, p53); cytoplasmic (PDGFR, COX-2); membrane (Her-2 / neu, EGFR). For the colon carcinoma, the targets were: nuclear (Ki-67, p53); cytoplasmic (COX-2, CEA); membrane (Her-2 / neu, EGFR). For the lung carcinoma, the targets were: nuclear (Ki-67, TTF-1); cytoplasmic (CEA, cytokeratin 7); membrane (Her-2 / neu, EGFR).
Results: We report the successful creation of both epithelial recognition algorithms (ERAs) based on selective imaging properties (Gaussian kernel and elongation ratio) and specific recognition algorithms (SRAs) based on pixel colors (RGB) and gray scale intensities that can analyze virtual slides created by TMA scanning. These algorithms rapidly identify cancer cells in a background of stroma, filter out this non-cancer background, successfully compartmentalize the cancer cells into nucleus, cytoplasm and membrane and then accurately quantitate the degree of immunocytochemical staining. The algorithm-based measurements strongly correlate with the subjective measurements with the strength of correlation being greater for nuclear (0.9) than membrane (0.8) than cytoplasm (0.7) (Confidence Intervals: 0.59, 0.95). The algorithmic measurements exhibit no interobserver, intraobserver or fatigue variability which is observed for the subjective interpretations.
Conclusions: These digital algorithms can presently be used to interpret virtual slides of TMAs in a more objective, more quantitative and more reproducible manner. Because of the increasing use of TMAs in both biomarker discovery and validation predicted to occur in the near future, this automatic interpretation is a prerequisite for making TMAs truly high throughput.
Category: Pathobiology

Monday, March 26, 2007 9:15 AM

Platform Session: Section G 1, Monday Morning