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

[1588] Use and Validation of ERA and FOV Algorithms on Virtual Slides To Guide TMA Construction

SH Barsky, L Gentchev, AS Basu, RE Jimenez, 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 interpretation. Because of the increasing demand for TMAs predicted to occur over the next decade, we felt it necessary to investigate whether their construction could be, at least partially, automated.
Design: 100 glass slides each of breast, colon and lung cancer were made into virtual slides. Image acquisition utilized scanners capable of producing images with a resolution of 20 pixels /10. We created both epithelial recognition algorithms (ERAs) and specific field of view (FOV) recognition algorithms that could analyze virtual slide images to select the areas richest in cancer cells. The glass slides were also used as a template to make corresponding TMAs in the traditional manner. We then compared the areas selected by our algorithms on the whole slides with the corresponding 1 mm² cores of the TMAs which had been manually constructed and measured their tumor cell densities (epithelial percentages).
Results: Our imaging algorithms successfully divided a virtual slide into a grid of hundreds of fields of view (FOV), each 1 mm², analyzed each field and identified cancer cells in a background of stroma and calculated the epithelial percentage in each field based on the ERAs. The ERAs in turn were based on applying the imaging principles of the Gaussian kernel and the elongation ratio. Those FOV with the highest epithelial percentages were mapped in terms of slide co-ordinates to be used for future TMA construction. The algorithm-based determinations were the same every time the algorithm was run on the virtual slide and therefore showed no interobserver, intraobserver or fatigue variability. Our comparative study showed that algorithm-driven selection of cancer areas were overall 25%-50% greater in cancer cell density than the cores of the TMAs which had been manually selected (p<.001). The overall algorithm-driven process took less than 15 minutes per slide to complete.
Conclusions: Our digital image algorithms can presently be used to guide the construction of TMAs or, in the future, to direct a robotic device to fully automate TMA production. Because of the increasing use of TMAs in both biomarker discovery and validation predicted to occur in the near future, this automation is a prerequisite for making TMAs truly high throughput.
Category: Techniques

Monday, March 26, 2007 11:45 AM

Platform Session: Section G 2, Monday Morning