Centromere detection of human metaphase chromosome images using a candidate based method
URL with Digital Object Identifier
Accurate detection of the human metaphase chromosome centromere is a critical element of cytogenetic diagnostic techniques, including chromosome enumeration, karyotyping and radiation biodosimetry. Existing centromere detection methods tends to perform poorly in the presence of irregular boundaries, shape variations and premature sister chromatid separation. We present a centromere detection algorithm that uses a novel contour partitioning technique to generate centromere candidates followed by a machine learning approach to select the best candidate that enhances the detection accuracy. The contour partitioning technique evaluates various combinations of salient points along the chromosome boundary using a novel feature set and is able to identify telomere regions as well as detect and correct for sister chromatid separation. This partitioning is used to generate a set of centromere candidates which are then evaluated based on a second set of proposed features. The proposed algorithm outperforms previously published algorithms and is shown to do so with a larger set of chromosome images. A highlight of the proposed algorithm is the ability to rank this set of centromere candidates and create a centromere confidence metric which may be used in post-detection analysis. When tested with a larger metaphase chromosome database consisting of 1400 chromosomes collected from 40 metaphase cell images, the proposed algorithm was able to accurately localize 1220 centromere locations yielding a detection accuracy of 87%.
metaphase chromosome, centromere detection, algorithm
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Citation of this paper:
Subasinghe A, Samarabandu J, Li Y et al. Centromere detection of human metaphase chromosome images using a candidate based method [version 1; peer review: 2 approved with reservations]. F1000Research 2016, 5:1565 (https://doi.org/10.12688/f1000research.9075.1)