Automated nuclei clump splitting by combining local concavity orientation and graph partitioning

Abstract

Automated clump decomposition is essential for single cell based analysis of fluorescent microscopy images. This paper presents a new method for automatically splitting clumps of cell nuclei in fluorescence microscopy images. Nuclei are first segmented using histogram concavity analysis. Clumps of nuclei are detected by fitting an ellipse to the segmented objects and examining objects where the fitted ellipse does not overlap accurately with the segmented object. These clumps are then further processed to find concave points on the object boundaries. The orientation of the detected concavities is subsequently calculated based on the local shape of the object border. Finally, a graph segmentation based approach is used to pair concavities that represent best candidates for splitting touching nuclei based on properties derived from the local concavity properties. This approach was validated by manual inspection and has shown promising results in the high throughput analysis of HeLa cell images.