Segmentation of cell nuclei using intensity-based model fitting and sequential convex programming

Leonid Kostrykin, Christoph Schnörr, Karl Rohr

Abstract: We introduce a convex model-based approach for the segmentation of cell nuclei, which exploits both shape and intensity information. The model is directly fitted to the image intensities. Previous shape-based approaches either are not globally optimal or require prior binarization of an image. Our approach relies on a fast second-order optimization scheme to solve a sequence of convex programs and estimate the globally optimal solution based on the image intensities. Model fitting is performed within image regions which are determined by exploiting the local image structure. We evaluated our approach using fluorescence microscopy images of two different cell types and performed a quantitative comparison with previous methods.