CLASSIFIED REGION GROWING FOR 3D SEGMENTATION OF PACKED NUCLEI

Automated 3D image segmentation and classification of biological structures is a critical task in modern cellular and developmental biology. Furthermore new emerging acquisition systems, like light-sheet microscopy, permit to observe whole embryo or developing cells in 4D, allowing us to better understand the spatial organization of tissues and cells. Numerous algorithms have been developed for 3D segmentation of cell nuclei, however when the cells are packed, classical methods usually fail. We present a new alternative for segmentation and classification by merging them into one classified region-growing algorithm. The combination of region growing and machine learning enabled us to both segment touching nuclei, and also classify them, even if their shape is changing in a dynamic context.