Fast and Automatic Nuclei Extraction from Cluttered Images in 3d Confocal Microscopy

We describe here a framework for robust and automated extraction of nuclei from cluttered 3D images, enabling the quantification of parameters including number of cells, number of mitoses and mitotic orientation. The proposed framework combines novel PDE-based image processing techniques for image filtering and analysis. Starting from a 2-channel staining of nuclei and membrane markers of the tissue (in our case the embryonic mouse heart), the membrane image is first filtered using anisotropic diffusion filtering [1], while the nuclei image is smoothed via a combination of median and non-linear filters [2]. The filtered membrane image is then subtracted from the filtered nuclei image to enhance the distinction of densely packed nuclei areas. A hierarchical thresholding method is then applied to obtain the approximate location and shape of each nucleus [3]. Finally, the boundary of each predetected nucleus is adjusted using a 3D active contours-based approach called Active Meshes [4], achieving fast and accurate segmentation via region-based energy minimisation.