Damir Sudar*, Steve Lockett, Mark de Kanter[2], Kasper Ligtenberg[2], Gus van der Feltz[2], Dan Pinke[1], Joe Gray[1]
Resource for Molecular Cytogenetics; MS 74-157; Lawrence Berkeley National Laboratory; 1 Cyclotron Road; Berkeley, CA 94720.
A number of important molecular cytogenetic analyses depend on the analysis of images of metaphase chromosomes. This includes applications such as DNA probe mapping relative to chromosome bands or as a fractional location along the length of the chromosome [1], CGH chromosome ratio profile measurements [2], translocation detection, and conventional banding-based karyotyping. These analyses rely on the accurate segrnentation of chromosomes from the background in the image, decomposition of clusters of chromosomes, and proper determination of chromosome borders. Locations of features such as bands in karyotyping, DNA probes in probe mapping, increases and decreases in CGH analysis, and breakpoints in translocation analysis need to be calculated with high accuracy which is complicated by chromosome bending and differential contraction. We have developed algorithms specifically suited for the segmentation of metaphase chromosomes in fluorescently labeled images and for the accurate determination of location of features along their length.
Automatic segmentation is performed in two stages: presegmentation of the metaphase image, and detection and decomposition of the clusters in the image. In the presegmentation, two parallel strategies are employed: one using operations from (binary) mathematical morphology, the other using a digital Laplace filter for edge detection. These two approaches are combined to get a segmentation along the most likely edges. Clusters of chromosomes are detected based on morphology and size. They are decomposed using on a rule-based decision system between likely cut-points on the contour of the cluster. Likely cut-points are determined from the contour curvature and skeleton branching. Segmentation results of 94% of the chromosomes in a metaphase were achieved for DAPI stained human metaphase spreads.
Mapping of locations along the length of individual segmented chromosomes was achieved by calculating the skeleton of the segmentation mask, converting the skeleton to a smoothed piece-wise linear line description (medial axis), extension to the telomeres of the chromosome, and the extraction of an integrated profile over the width at unit-step locations along the medial axis.
In order to correct for errors in the telomere location, differential stretching of the chromosomes, and differences in chromosome lengths we implemented a recursive warping algorithm of the profiles which converts a profile to a template based on 'matchable' locations along the profiles such as centromeres, telomeres, and chromosome bands. This allows the combination of profiles from multiple metaphase by averaging which is essential for analysis methods such as CGH.
This work was funded by the US DOE contract DEAC0376SF00098.
[1] University of California, San Francisco, CA.
[2] Delft University of Technology, Delft, The Netherlands.
* Corresponding author
[1] Mascio L.N., Verbeek P.W., Sudar D., Kuo W-L, Gray J.W., Semiautomated DNA Probe Mapping Using Digital Imaging Microscopy: I. System Development. Cytometry 19:51-59, 1995
[2] Piper J., Rutovitz D., Sudar D., Kallioniemi A., Kallioniemi O-P, Waldman F., Gray J.W., Pinkel D. Computer Image Analysis of Comparative Genomic Hybridization. Cytometry 19:10-26, 1995