Processing And Analysis Of Methaphase And Chromosome Images Ekaterina Detcheva

PROCESSING AND ANALYSIS OF METHAPHASE AND CHROMOSOME IMAGES Ekaterina Detcheva Institute of Mathematics and Informatics, Bulgarian Academy of Sciences, Acad. G. Bonchev St., bl. 8, 1113 Sofia, Bulgaria E-mail: [email protected] The paper describes the realization of procedures for processing and analysis of metaphase and chromosome images. Image processing procedures include preprocessing and image segmentation. Analyzing procedures extract methaphase and chromosome features. A description of these procedures and corresponding data structures is given. 1. Introduction Chromosome analysis intelligent system (CAIS) is developed to perform chromosome analysis and classification. Methaphase and chromosome image processing and analysis are based on analyzing the gray-level images of biological objects on a microscope slide. The images are composed of biological objects nearly well spread and distinguishable from the background. Image processing methods include preprocessing and image segmentation. Analysis methods extract methaphase and chromosome features to form a symbolic image description. All these methods are applied on large data structures like pixel arrays. A high efficiency and a lower degree of flexibility are required to handle such large amounts of data. To be successfully realized these methods should be implemented as appropriate efficient procedures. This problem requires special domainoriented knowledge about image processing and chromosome classification. The present paper describes the realization of metaphase and chromosome image processing and analysis procedures organized in classes of objects [1,2]. 2. Image processing procedures This section contains a short description of procedures for methaphase and chromosome image processing. 2.1 Loading an image from a file The system is working on 256-grey-level images in the following graphic file formats:

Windows Bitmaps (BMP), Tagged Image File Format (TIFF) and Raw Image Format (PIC). Images can be viewed in a 25%, 50%, 100%, 200% or 400% scale (Figure 1).

Figure 1: Methaphase image 2.2 Pre-processing Methaphase image can be interactively edited if there are some touching chromosomes, artifacts etc. Methaphase smoothing is made using median filtering. 2.3 Chromosome isolation Object isolation is performed by a recursive procedure analyzing 3x3 window with a center the current pixel with gray value greater then the threshold. The result of isolation is displayed on the screen. Each isolated object is drawn in a separate rectangle. All objects are sorted in decreasing order of their area (Figure 2). 2.4 Chromosome editing Some of isolated objects can be overlapped. They can be isolated interactively by the operator (Figure 3). 2.5 Chromosome centromere location Initially LaPlassian filtering using 3x3 convolution on the current window with the kernel below performs object contour detection: 0 -1 0

-1 4 -1

0 -1 0

Figure 2: Isolated chromosomes

Figure 3: Object editing

Figure 4: Centromere location Then chromosome centromere is located using the “closest approach”[3]. The two opposite contour points with longest distance are determined first. They define the left

and the right contour part of the chromosome contour. The two opposite contour points with the shortest distance define the centromere position. The location of the centromere is in the middle of the segment connecting the two contour points (Figure 4). In order not to misplace the centromere at the ends of the chromosome the top and the bottom points are ignored. This approach can be successfully applied also to bent chromosomes. It is independent from the method of chromosome painting. 2.6 Chromosome length and centromeric index Next the lengths of the two chromosome arms are measured. The length of each arm is calculated as the longest distance between the centromere position and each contour point of the arm. The total chromosome length is calculated as a sum of the two arms’ lengths. Finally centromere index is computed as a ratio between the length of the short arm and the length of the chromosome. 2.7 Centromere position editing The chromosome position of the centromere is displayed on the screen and the operator can interactively make corrections of it if it is misplaced. 2.8 Chromosome upright setting The chromosomes are oriented with the short arm pointing in the Y-direction of the scanning gird (Figure 5). For this purpose the angle between the line connecting the two arms end points and the X-direction of the scanning gird is calculated. The angle is given by: x1 – x2 α = atan  y1 – y2 where (x1,y1) and (x2,y2) are the coordinates of the two end points of the chromosome. If y1 = y2, the angle is given by: π α =  when x1 > x2 2

π α = -  when x1 < x2 2

If y1 - y2 < 0, the angle is: α = α + π.

Figure 5: Chromosome upright setting

2.9 Bands detection To detect the bands within the chromosome a convolution with a two-dimensional LaPlace filtering is used [3]. The coefficients are given by: 2 -1 -2 -1 2

2 -1 -2 -1 2

2 -1 -2 -1 2

2 -1 -2 -1 2

2 -1 -2 -1 2

The points having a positive value after this procedure belong to the band pattern of the chromosome. 2.10 Bands description For bands matching a so-called band vector is constructed. Black and white points in each row of the chromosome image are counted. If the black points are more then the white current position of the band vector is set to 0 (black). Otherwise it is set to 255 (white). In this way a specific band pattern of each chromosome is built (Figure 6). 2.11 Bands matching A special function is developed to count the number of matching points in each chromosome band vector and previously given landmark pattern (Figure 6). The landmark patterns are taken from “Idiogram Album: Human”, 1994, David Adler.

Figure 6: Bands matching 3. Data description The processing and analyzing of methaphase and chromosome images is based on analyzing the gray-level images of biological objects on a microscope slide. Standard procedures for image processing such as median filtering, negation, segmentation, contour detection, etc., defined as class methods and corresponding data like matrix representing the digitized image, the matrix sizes, gray level thresholds are organized in object class named IMAGE. The chromosome images are gray-level microscope images and the data and procedures of this class can be used for the chromosome image processing and analysis. Certain specific data and procedures are also needed such as chromosome position, chromosome area, length and centromere index, chromosome arrangement by area. These data and procedures are organized into the class CHROMOSOME that is a subclass of the IMAGE class. The metaphase images are gray-level microscope images. The data and the procedures of the class IMAGE can be used for processing the metaphase images. In addition certain specific data and procedures are needed for the processing of the metaphase images, e.g. number of isolated chromosomes, isolated chromosome images, editing of touching and overlapping chromosomes, bands analysis. These data and procedures are organized into container-class of objects named METAPHASE. The described above classes of objects are used for the chromosome image processing and analysis. The hierarchy of these classes is shown on the figure 7 [4]:

Figure 7: Main classes For bands matching an array named ClassLandmarks is built. It contains 24 objects named Landmarks. Every object is a dynamic array of objects of class Band (Figure 8). It contains: − −

Band pointers Number of bands

Bands characteristics for each individual chromosome landmark are: − − − − −

Arm Position Color Start End

Figure 8: Bands data representation 4. Conclusion The purpose of this paper is to describe an approach to the mathaphase image processing and analysis. Currently most systems [5] are based on extremely complicated methods

for centromere location, chromosome measurement and band description. The main advantage of presented above methods is in their simplicity. Another point of interest is that object isolation, centromere detection and chromosome length calculation follow the way of visual karyotyping. The system CAIS is based upon the IBM PC computers and is written in C++ for Windows. The implementation of the system is under way. The classification algorithms are being built. The system CAIS can be extended by addition of new classes of objects that represent other sort of pre-processing, analysis and classification. Acknowledgments. The system is developed at the Artificial Intelligence Department of the Institute of Mathematics and Informatics, Bulgarian Academy of Sciences. The Project is partially sponsored by the Ministry of Education, Science and Technology under contract No I-531. References. [1]

Decheva E. Object-oriented approach to the design of CAIS - an intelligent system for chromosome analysis and classification. Mathematics and Education in Mathematics. Proc. of the 21st Spring Conference of the Union of Bulgarian Mathematicians, Sofia, 1992, pp. 226-230.

[2]

Piper J., R. Balldok, S. Towers, D. Rutovitz. Towards a knowledge-based chromosome analysis system. In: Automation of Cytogenetics. C Lundsteen, J. Piper (eds.). Springer-Verlag, Berlin, 1989, pp. 275-290.

[3]

Ten Kate T.K. Design and implementation of an interactive karyotypingprogram in C on a VICOM digital image processor. I2 Report. Systems and signals group,Section pattern recognition, Delft University if Technology, 1985, p. 78.

[4]

Detcheva E. CAIS – an intelligent system for chromosome analysis and application. Int. Workshop “Artificial Intelligence and the Humanities, Sozopol, September 1996, pp. 49-54.

[5]

Lundsteen C., J.Piper (eds.). Automation of Cytogenetics. Springer, Berlin, 1989, p. 316.