Direct classification of human G-banded chromosome images using support vector machines

Automatic classification of chromosome images used in karyotyping has been of interest for many years. Regardless of the efforts put into this field, due to the complexity of the matter, still the functional accuracy of current automated systems is much lower than a human operator. Since the interdiction of SVM and its proven efficacy in pattern recognition both in theory and application, we decided to test it’s efficacy on G-banded chromosomal images. The results were significantly more favorable. The recognition rate in chromosomal subgroups averaged at 95.9%. Furthermore, alongside this study an unmatched database of chromosomal images with about 42000 items was created which can be used as a reference database for further research in this field.