Automatic Segmentation of the Whole G-band Chromosome Images Based on Mask R-CNN and Geometric Features

Chromosome segmentation is an important task in the human chromosome analysis process and it plays a significant role in chromosome classification and karyotyping. In this paper, automatic segmentation method of the whole G-band chromosome images based on Mask R-CNN and geometric features is proposed to accurately segment the whole image. Firstly, preliminary segmentation based on Mask R-CNN is proposed to obtain all chromosome clusters. Using the size of the chromosome to get real chromosome clusters. In order to correctly disentangle touching and overlapping chromosomes, a further segmentation method based on the geometry features of the chromosome is suggested. We have applied our method to the database which contains 1148 metaphase chromosome images and 4568 touching and overlapping chromosomes images. Experimental results show that our algorithm brings an encouraging performance compared to other methods.