Band features as classification measures for G-banded chromosome analysis

Modern automatic and semiautomatic karyotyping systems employ algorithms that use chromosome length and centromeric index as well as other intact chromosome measures. These measures offer correct classification rates near 95%. An algorithm is presented that utilizes local dark band features and position (position from one end of the chromosome, band-width, band-height above light band background, integrated optical density above light band background, and a shape feature) and is based on maximum likelihood of the multivariate normal distribution for the feature vector. The algorithm was tested on two data sets: 179 metaphases from C. Lundsteen at the Rigshospitalet, Copenhagen, and 50 metaphases from The University of Texas M. D. Anderson Cancer Center. The Copenhagen set achieved an overall correct classification rate of 94.6% when classifying itself, a rate comparable to other algorithms. This classifier relies on local band features rather than global chromosome characteristics and is therefore directly extensible to metaphase and prophase chromosome subsegments and to structural abnormalities.