(Authors): Resource for Molecular Cytogenetics. MS 74/157. Ernest Orland Lawrence Berkeley National Laboratory. University of California
Segmentation of individual cell nuclei from within intact tissue specimens is an important basic capability for studying heterogeneous cell populations and the spatial organization of the cells. For quantitatively accurate measurements on each nucleus, the nuclei must be intact, which requires the use of thick (>20 microns) specimens, 3D (confocal) image acquisition and 3D segmentation algorithms.
In this study we developed a 3D segmentation approach that combines the accuracy of the human visual system with the computational power of image analysis algorithms. The approach is as follows: first automatic algorithms separate the 3D image into nuclear regions and background. Then using a graphical display that shows the surface rendering of each nuclear region intersected with 2D slices from the original 3D image, the analyst visually classifies the nuclear regions as either as individual nuclei or clusters of multiple nuclei. Next, an automatic, distance transform based algorithm divides the clusters into smaller objects, which are again classified by the analyst. Once no more cluster remain, the user can indicate which objects should be joined to form individual nuclei and which are debris.
The system has been assessed using 3D images of DNA labeled C-elegans embryos, benign breast cancer tissue and tumor breast cancer cells cultivated in mice The results show 98% of nuclei correctly segmented in embryos, and 90% in the other samples.