Getting poster data...
Shantanu SIngh, Raghu Machiraju, Thierry Pecot, Gustavo Leone (395 Dreese Laboratories, 2015 Neil Avenue, Columbus, OH 43016)We are building a framework for large scale phenotyping of nuclei in cellular stroma. The framework can be used to conduct both, hypothesis-driven studies, such as testing whether nuclear morphologies change across different genotypes or stages of cancer, as well as to conduct exploratory studies, such as discovering novel nuclear phenotypes in the microenvironment. Mouse models of cancer with various of combinations of tumor-suppressor knock-out (PTEN) and oncogene over-expression in fibroblasts form the basis of the study. In these mice, transgenic fluorescence is expressed in salient cell types including fibroblasts. An image analysis workflow is constructed to segment cell nuclei and the 3D nuclear morpholology is modeled using a point distribution model followed by a spherical harmonics decomposition. A low dimensional space using PCA is used classify nuclei across cell types. Further, using metric learning it is possible to use nuclear texture and spatial context to distinguish between similar-shapes. Finally to learn subtle changes, permutation testing is conducted and distributional differences are determined from large populations of cells.