For developing an automated system for analysis of chromosomes and their classification into pairs, certain uniquely defined features need to be taken into consideration. The system is developed for high resolution image formats like .bmp, .tiff and .jpg. Also, the system will support images having non-overlapping straight chromosomes. The objective of the proposed system is to develop a cost-effective alternative for commercially available softwares.
To develop an automated system for chromosome orientation and pairing, the images in gray scale format are processed and enhanced. A cluster of chromosomes along with slide artifacts need to be separated using intelligent algorithm. The target is to obtain an output image with well-defined vertically aligned chromosomes in the order of length, centromere position and band patterns. The first phase starts with segmentation of objects for future extraction and analysis. Global threshold value for the image accounts for segmentation of objects from the unwanted background pixels. Thereafter, extraction is performed to administer individual object properties. Each chromosome is extracted the way it is present in the native image. Hence, for karyotyping proper rotation is necessary. Edge linking techniques like local processing and global processing can be used for detecting edge lines. However, the former implementation largely depends on pixel similarity measures like magnitude and angle of gradient vector. Thus the space and time complexity increases for processing each and every pixel. A global processing technique such as Hough Transform on the other hand is a way of finding edge points in an image that lie along a straight line, curves and other structures if their parametric equation is known. It is resistant to noise and can produce multiple instances of a model in a single pass, which proves to be a favorable choice. While vertical rotation takes place, it has to be taken care that the chromosome is properly placed. This implies that the short arm (p-arm for petite) should be positioned above the longer arm (q-arm for queue). This can be achieved by computing the distance of the centromere from both the ends using Medial Axis Transformation. Following phase deals with merging the individual chromosomes into a karyotype image in the order of decreasing length so as to conform to the ideogram rules.
For band pattern matching, which is the final stage in automatic karyotyping, a density profile signal has to be computed which produces a 1D vector, the values of which contain the median values of the pixels orthogonal to the skeleton or medial axis. 2D Laplace filtering is then used for detecting hills and valleys in gray values which is analogous to the band patterns in chromosomes.
Automatic Karyotyping 