Splitting of cell clumps in cytokinesis-blocked micronucleus images: application to improve the recognition ability of binucleated lymphocytes

Abstract

The cytokinesis-blocked micronucleus (CBMN) test in human peripheral blood lymphocytes is an extensively used method for biomonitoring, mutagenicity testing, and especially radiation biodosimeter. Automated analysis of CBMN by image processing can provide faster and more reliable results with minimized subjective micronucleus (MN) identification than the time-consuming and tiresome manual scoring. Splitting of the clumps, i.e. overlapping lymphocytes, overlapping nuclei, and combined overlapping of nucleus and MN, remains an unsolved problem in the automated analysis of CBMN images, because ignoring them will dramatically decrease the recognition ability of binucleated cells and directly affect the statistical validation of MN frequency. We present a novel algorithm for splitting these clumps in CBMN images based on improved watershed transform. Experimental results show that this algorithm has succeeded in correctly splitting the clumps composed of nuclei or lymphocytes as well as combined clumps of MN and nucleus. This presented algorithm is valid for the splitting of clumps in CBMN images, and may also be adopted for other clumps where splitting of overlapping regions is required.