Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Electrical and Computer Engineering


Dr. Jagath Samarabandu


Development of an effective human metaphase chromosome analysis algorithm can optimize expert time usage by increasing the efficiency of many clinical diagnosis processes. Although many methods exist in the literature, they are only applicable for limited morphological variations and are specific to the staining method used during cell preparation. They are also highly influenced by irregular chromosome boundaries as well as the presence of artifacts such as premature sister chromatid separation.

Therefore an algorithm is proposed in this research which can operate with any morphological variation of the chromosome across images from multiple staining methods. The proposed algorithm is capable of calculating the segmentation outline, the centerline (which gives the chromosome length), partitioning of the telomere regions and the centromere location of a given chromosome. The algorithm also detects and corrects for the sister chromatid separation artifact in metaphase cell images. A metric termed the Candidate Based Centromere Confidence (CBCC) is proposed to accompany each centromere detection result of the proposed method, giving an indication of the confidence the algorithm has on a given localization.

The proposed method was first tested for the ability of calculating an accurate width profile against a centerline based method [1] using 226 chromosomes. A statistical analysis of the centromere detection error values proved that the proposed method can accurately locate centromere locations with statistical significance. Furthermore, the proposed method performed more consistently across different staining methods in comparison to the centerline based approach. When tested with a larger data set of 1400 chromosomes collected from a set of DAPI (4',6-diamidino-2-phenylindole) and Giemsa stained cell images, the proposed candidate based centromere detection algorithm was able to accurately localize 1220 centromere locations yielding a detection accuracy of 87%.