An automated detection and morphological classification of numerical abnormalities in human chromosomes

Cytogenetic is a branch of genetics that is concerned with the study of the structure and function of the cell, especially the chromosomes. The chromosomal identification is of prime importance to geneticist for diagnosing various abnormalities. The existing system is developed to classify the chromosomes based on pixel distribution, centromere index and band patterns using artificial neural network techniques. The accuracy of classification is lowered particularly in sub group `C´. In this paper we propose a technique where the input images of the unpaired well spread chromosomes are obtained from the electron microscope. Initially noise is removed and edges are detected. Then, each object is extracted from the input image, rotated to align vertically and cropped. Then, the features of each chromosome like major axis length, Area and histogram are analysed and sorted in descending order to perform classification. Then based on the number of objects, the numerical abnormality like monosomy and trisomy are detected. Thus the system is fully automated for well-spread images and semi-automated for images with overlapped chromosomes.