Poster Session

Background: Image analysis algorithms, coupled with maturing digital whole slide imaging (WSI) technology, holds promise to provide tools for morphometric quantitation in surgical pathology.However, implementation of such strategies will require development and optimization of pattern recognition algorithms adaptable to diseases showing complex architectural features and cytologic atypias.One such example is urothelial carcinoma (UC), of which the aggressive micropapillary variant (MPUC), an aggressive variant of UC which is frequently under-recognized causing diagnostic diffi culties.Herein, we demonstrate the potential of a recently described pattern recognition algorithm and its application to this challenging use case.Methods: We have recently reported SIVQ (Spatially Invariant Vector Quantization), an algorithm that uses ring vector predicates for pattern recognition (Hipp and Cheng, 2010).However, the relative contributions of key SIVQ ring parameters have not been fully characterized.Consequently, we systematically tested SIVQ ring parameters for detection of micropapillary nests in fi elds of a classic case of MPUC by comparing pattern match quality scores at pixels inside and outside of a pathologist-determined "ground truth", using the receiver operating characteristic (ROC) curve.Results: To standardize ring parameter optimization, we tested various ring diameters, number of subrings, and inter-ring rotational "wobble" angles.First,