Image Analysis Techniques for Segmentation, Classification and Presentation of High-Resolution Prometaphase Chromosome
The extraction, analysis, and presentation of integrated optical density (IOD) profiles from banded prometaphase chromosome material is an increasingly important clinical method for the detection and analysis of structural aberrations. Particular translocations, duplications, or deletions of genetic material have been shown to be implicated in specific mental or physical disorders. Previously, this clinical task was carried out by a laborious manual procedure with little or no automation. This thesis describes an investigation into digital image analysis tech niques for automation in cytogenetics. More specifically, it presents a number of image analysis methods implemented in a clinically useful computer system for studying the banding structure of stained human chromosomes. The computer system consists of a microscope image digiti sation unit based on a high-resolution (1320 X 1035 pixels, 12 bits per pixel) digital CCD cam era and an image analysis and presentation unit based on a standard graphics workstation. The more precise and objective indications of genetic abnormality.
A new method of chromosome classification is proposed based on a conventional Bayesian classifier presented with sampled banding profile values as features. This straightforward method of classification which reflects the human cytogeneticist's approach appears to have been neglected in the past, but gives good results. Using approximately 20-30 profile values (features) along the chromosome centreline, the classification result is better than most of the previously reported classifiers.
The same classifier is then exploited to solve two related problems. The first is the segmen tation task of identifying which sections belong to which chromosome when two or more chro mosomes overlap. This is by no means obvious with material in which chromosomes can be severely bent or twisted. The second is the axial alignment problem of matching corresponding bands in chromosomes presented to the cytogeneticist for comparison. Alignment is based on the intuitive concept of piece-wise correlation, designed to handle the alignment of high-reso lution profiles with detailed differences in chromosome band structure, and using the sample classifier to evaluate the goodness of fit.
The thesis also presents a clinical experiment in which image quality of microscope ocular images, photographs from microscope built-in camera, digital images from standard and high- resolution cameras, and sublimation printer hardcopies was compared. It was shown that the developed high-resolution digitisation unit was superior to standard cameras as judged by number of visible chromosome bands.
digital image analysis
high-resolution CCD camera