Automated detection of cut-points for disentangling overlapping chromosomes

Automating chromosome classification and segmentation has been a major challenge in automated karyotyping especially due to overlapping chromosomes. The earlier reported methods for disentangling the chromosome overlaps have limited success as they are sensitive to scale variations, computationally complex, use only color information in case of multispectral imaging and most of them are limited to separation of single overlap of two chromosomes in a cluster. This work proposes a heuristic approach to locate the cut-points on an overlap and thus addresses the first step towards automated extrication of overlapping chromosomes from the metaphase image. The proposed simple, but novel and robust algorithm, to identify the appropriate cut points is based on computational geometry of the pixels on the boundary of the cluster. Contribution of this work is in the ability of the algorithm to successfully identify the correct cut points in multiple overlaps with multiple chromosomes. Moreover, as the proposed algorithm is independent of shape and color information, it also applies to Multicolour Fluorescence in-situ Hybridization (MFISH) metaphase chromosomes, System performance was tested and analyzed using 40 real images giving an overall average accuracy of 98% for correct identification of cut points in 2, 3 and 4 overlapping chromosomes.