Automated pairing of human chromosomes applying gradient profile and similarity matching algorithm

One of the most important tasks in cytogenetics is chromosome analysis, with which cytogeneticists can diagnose whether there are structural or numerical chromosome aberrations. In this paper, a feasible way to deal with computer-aided pairing of metaphase chromosomes is presented. From pretreating images to pairing chromosomes, feature extraction and pairing methods are described in detail. As an important feature, density profiles are extracted from banding patterns and smoothed by both median and mean filters. In these cases, density profiles are adjusted by the method proposed in this paper. Gradient profiling, a new feature attained by transforming the density profile along with the length and centromeric index, is used in our chromosome pairing procedure. Chromosomes are divided into 23 pairs by comparing their similarity, which is numeralized as the cosine value of gradient profiles. Compared with density profiling, gradient profiling improves matching accuracy by 12.73%. Furthermore, classification accuracy of chromosomes in male and female cells is 91.3% and 100% respectively. Experimental results indicate that the proposed gradient profiling method can serve for chromosomes pairing.

► We proposed a scheme for automated chromosomes pairing including images pretreating, feature extractions and pairing method.
► We proposed a new feature, gradient profile, which could better reflect the characteristics of the banding patterns.
► Comparing with density profile, gradient profile improves the matching accuracy with 12.73%.