Expectation Maximization Segmentation Algorithm for Classification of Human Genome Image

Abstract

Chromosomes are cellular structures which carry the genetic material. A chromosome is composed of a single circular or linear DNA molecule. The cell details of every individual is found in genome which contains the DNA genetic blueprint. The partitioning and categorization of chromosome has to be automated by standard algorithms for easy diagnosis of certain diseases. The structural and numerical anomalies in the genes that occur to the future generation can be predicted through the analysis of the various characteristics of the chromosomes. Karyotyping indicates the display of the chromosomes of a cell by arranging them in a specific and distinct fashion which will simplify the chromosome analysis. The multispectral staining techniques adopted in MFISH offers classification of human genome by assigning different colors to different chromosomes that ease the determination of structural and numerical aberrations. The important step in multispectral MFISH image karyotyping is segmentation of DAPI images. In this paper, Expectation maximization algorithm for M-FISH segmentation is presented. The Expectation Maximization segmentation algorithm reveals improved performance in segmentation when an analogy is made with watershed segmentation method for 30 sets of images taken from ADIR dataset of MFISH images. After segmentation, chromosomes ar classified using K means algorithm and an overall quality of 91.68% is reported.