Automated homologue matching of human G-banded chromosomes

Two sets of human G-banded chromosomes were employed to test a computer algorithm for homologue matching. One set was produced at the Rigshospitalet, Copenhagen, and the other at the University of Texas M. D. Anderson Hospital. Employing a cross-correlation measure to select candidate homologue mates for each chromosome resulted in correct matches in 91.1% of the cases in the Anderson set and 93.1% in the Denmark set. To identify each chromosome and to give a measure of classification accuracy when performed by humans, five cytogeneticists were asked to independently karyotype the 49 cells in the Anderson set. Agreement between two cytogeneticists was measured by the kappa statistic. If all chromosomes that could not be identified by any given cytogeneticist were removed from the comparison, kappa values were found to be in the vicinity of 0.95. With such unidentifiable chromosomes included as a separate class, the kappa values were closer to 0.89.