15-03-2010, 08:40 PM
[attachment=2655]
A PAPER
ON
CHROMOSOME IMAGE ENHANCEMENT AND IDENTIFICATION
Presented By:
Mr. V. Subba Reddy,
ME (Digital Systems),
Abstract:-
Image processing has been used in almost all the areas such as remote sensing, military, biomedical, industrial applications. The world is growing more crowded all the time, with a fragile environment that needs protection. Natural resources are becoming increasingly precious and hard to find.
This paper is mainly focused on chromosome image enhancement and future extraction based on karyotyping. The features computed and used for analysis include: centromeric index, chromosome size and banding pattern features including total number of bands. A method that is discussed is enhance the image by applying filtering technique and finds the future of centromere detection. This paper is covered the procedure of enhancement and detection of centromere and presented some of results. The results are obtained by different algorithms using MATLAB.
1. Introduction:-
In the past few years, immense improvement was obtained in the field of content based image retrieval (CBIR); nevertheless, existing systems still fail when applied to medical image databases. Simple feature extraction algorithms that operate on the entire image for characterization of color, texture or shape cannot be related to descriptive semantics of medical knowledge that is extracted from images by human experts.
Content based image retrieval (CBIR) system is required to effectively and efficiently use information from these image repositories. Such a system helps user (even though unfamiliar with the database) retrieve relevant images based on their contents. Application areas in which CBIR principles activity are numerous diverse:
Art galleries museum management, architectural and engineering design, interior design remote sensing and management of earth resources geographic information systems scientific database management weather forecasting retailing, fabric and fashion design, trademark and copyright database management, law enforcement and criminal investigation, picture archiving and communication systems.
The most important aspect of CBIR is Feature extraction. The features must efficiently describe the most important information conveyed to users in images. In a broad sense features may include both text based description (keywords annotations etc.) and visual features (color, texture, shape, spatial relationship etc.). Since text based feature extraction is already a well-defined field within information retrieval research communities. Visual features can be classified into general features and domain specific features. The formal include color, texture and shape features while the later are application dependent and may include, for example human faces and fingerprints, medical image.
2. Image processing in medical applications:-
Image processing has been used in almost all the areas such as remote sensing, military, biomedical, industrial applications. The world is growing more crowded all the time, with a fragile environment that needs protection. Natural resources are becoming increasingly precious and hard to find. Remote sensing images are obtained from satellite and aircraft sensors to study the natural resources available on the earth surface. Space probes are flown to other planets such as Mars, Venus, and Jupiter etc., to get the images of these planets. Various types of satellites are deputed to get high resolution reconnaissance images for military, applications. All these images require Image Processing to increase contrast to get the information by suppressing the noise signals arising out of platform motion, atmosphere and sensor non-linearities. [6]
With the advent of Image Processing techniques, network processing facilities and high resolution and high speed display work stations, doctors all over the world are able to interpret images of patients transmitted from remote areas for clinical diagnosis. Micro surgery is becoming popular due to the availability of Image Processing techniques to display human parts on 3D display systems. [6]
Imaging has become an essential component in many fields of medical and laboratory research and clinical practice. Biologists study cells and generate 3D confocal microscopy datasets, virologists generate3D reconstructions of viruses from micrographs, radiologists identify and quantify tumors from MRI and C T scan, and neuroscientists detect regional metabolic brain activity from PET and functional MRI scan. The use of medical imaging for diagnostic purposes as well as for evaluation of pharmaceutical trials has increased exponentially in recent years. [6]
3. CHROMOSOME FEATURES
There are many chromosome features that have shown capability of identifying and delimiting chromosomes. Most chromosomes can be classified based on their relative length, centromeric index, and banding pattern in the context of the cell. From Fig.2.1 notice that chromosome length gradually declines following from class 1“22. With subtle variations among cells, such as preparation technique, size, and chromosome morphology, many classifiers use that information via normalized-chromosome length for the cell. Besides length, the centromeric index provides a significant amount of chromosome delimiting capability. The centromeric index is commonly defined as the ratio of the chromosome™s short arm (p-arm) to the total chromosome length. The centromere is usually located in a constricted region along the chromosome contour. Fig. 2.1 illustrates the unique positions of centromeres for chromosome classes 4, 7, 15, 20, and X (black arrows). Classes 13, 14, 15, 18, 21, 22, and Y contain acrocentric centromeres near the top of the chromosome™s short arm. The other chromosome classes, where the centromere is closer to the middle of the chromosome, are referred to as metacentric.
3.1 Karyotyping: -
Karyotyping refers to the visualization and classification of chromosomes found in metaphase spreads. The bellow fig 2.1 contains a normal female metaphase spread and its corresponding karyotype. Visually, the chromosome pairs, known as homologues, appear similar. Normal cells contain homologues for chromosome classes 1“22, the autosomes, and the sex chromosome X for a female or paired X and Y sex chromosomes for a male.
Feature extraction is performed for each chromosome found within the metaphase spread image. The features computed and used for analysis include: 1) centromeric index,
2) chromosome size ( area of chromosome and area of p-arm and q-arm band), 3) banding pattern features including total number of bands.
Fig 2.1 Karyotype image
The centromere location, which may occur at unique points along the chromosome, also provides delimiting capability. Thus, knowing the centromeric index delimits the set of classes to which the chromosome belongs. The banding pattern provides important information for chromosome classification. Each of the 24 classes possesses unique banding patterns. The band level often indicates the degree of condensation and length of chromosomes found within a metaphase spread and, therefore, affects certain features, such as length and banding pattern. Unique banding patterns for each class provide for several band features, which are useful for chromosome-classification purposes. One of the simplest is the density profile, the mean grey level along perpendicular lines to the medial axis of the chromosome, providing a gray-level representation of the banding pattern. Due to variations in preparation technique, sources of the cells, and image enhancement techniques are accompanying image acquisition, it is difficult to consistently obtain a generalizable banding pattern for gray level chromosome images. Fig. 2.2 shows several chromosome 17â„¢s in a composite image, illustrating the variation in size and banding pattern for those chromosomes. Incorporating this knowledge, Zimmerman et al. performed chromosome classification primarily relying on the number of bands. [1]
Fig 2.2 Image of chromosome 17â„¢s from various cells
3.2 Cenntromere Detction for Chromosome Identification
A) Processing Method:-
Figure 3.1 demonstrates a block diagram of the proposed automatic centromere locating system. In the following sections the function of each block is described in detail.
B) Producing the Binary Image: -
Since the aim is to look for a morphological feature of chromosome, the grayscale information of the image is not of interest at this stage. Therefore, producing a binary image will make the rest of the process easier. The chromosome images are typically
Input image of a single chromosome
Coordinates of centromere
Fig 3.1 The proposed block diagram for centromere finding
bimodal in other words; such an image includes an object over a uni-color background. Histogram of such image usually includes two peaks, one of which corresponds to the background and the other to the object. The preferred threshold separating the object and background-related peaks in the histogram of such an image is usually the gray level representing the global minimum located between the two peaks. For a robust identification of the two major peaks and the global minimum between them, it is necessary to filter out the small variations in the histogram. For this purpose, a first order Savitzky-Golay FIR smoothing filter with a window width of 5 is found to be effective. Savitzky-Golay filters are low-pass filters useful for smoothing data. This time-domain method of smoothing is based on least squares polynomial fitting across a moving window within the data. The method was originally designed to preserve the higher moments within time-domain spectral peak data. Since the background related peak is of many orders of the object related peak in magnitude, a median filter with a window size of 9 is also applied to the histogram to make the peaks visually comparable and easier to locate automatically. Figure 3.3 shows a typical chromosome number 16 and its original and filtered histograms.Due to the usually white background of the images, the background related peak in the histogram is always located close to the gray level 255. [2]
However, a little bit more care is needed for the determination of the object related peak. After a statistical analysis of the images in the data set, the following two rules are found to be valid for all cases: a) if the mean grayscale of an image is greater than 200 then the object related peak is located between 0 and 220; b) however, for darker images (mean grayscale less than200), the object related peak is located between 0 and 150. By locating the two peaks representing the background and the object in the histogram, the gray level coincident with the global minimum between the two peaks is defined next. Using this value as a threshold, all the pixels with gray level below this threshold are set to 1 (white) and the remaining pixels to 0 (black) producing the binary format of the input image.[2]
However, a little bit more care is needed for the determination of the object related peak. After a statistical analysis of the images in the data set, the following two rules are found to be valid for all cases: a) if the mean grayscale of an image is greater than 200 then the object related peak is located between 0 and 220; b) however, for darker images (mean grayscale less than200), the object related peak is located between 0 and 150. By locating the two peaks representing the background and the object in the histogram, the gray level coincident with the global minimum between the two peaks is defined next. Using this value as a threshold, all the pixels with gray level below this threshold are set to 1 (white) and the remaining pixels to 0 (black) producing the binary format of the input image.[2]
Fig 3.2 A typical chromosome automatic algorithm number 16 and its original
(upper plot and filtered (lower plot) Histograms
C) Computing the projection vectors:-In order to calculate the horizontal projection vector, simply the pixel values of each row are summed up in the binary image. Considering that the binary image includes only 1s (white pixels) and 0s (dark pixels), therefore, each element of the horizontal projection vector is equal to the number of white pixels (1s) in the corresponding row. Similarly, to calculate the vertical projection vector, the pixel values of each column are summed up in the binary image. Theoretically, these two orthogonal projection vectors contain all the morphological information of the chromosome and can be used for binary image reconstruction and/or for feature extraction. For example they can be used to calculate the extent of the chromosome (i.e. where the chromosome begins and ends on the image plane) or they can be used for locating the centromere of the chromosome, which is explained next.[2]
D) Locating the centromere:- As it was explained before, from the morphological point of view, the centromere is the narrowest part of the chromosome in its longitudinal direction. This produces a global minimum in the central part of the horizontal projection vector of the chromosome, due to the small number of the white pixels (1s) in the horizontal direction at this region (see Fig.3.3).
Fig. 3.3 Centromere detected chromosome
Therefore, centromere locating is just locating this global minimum in the central region of the horizontal projection vector. Fig.3.3 shows the horizontal projection vector of the typical chromosome number 16 of Fig.3.2 with its global minimum in the central region is marked with a circle. This corresponds to the centromere location which is marked with a white line over the chromosomeitself.[2]
4.CONCLUSION
There is no universal procedure for threshold selection that is guaranteed to work on all images. There are a variety of alternatives different thresholding techniques are there. One alternative is to use a threshold that is chosen independently of the image data. If it is known that one is dealing with very high-contrast images where the objects are very dark and the background is homogeneous and very light.
A simple, yet very effective algorithm for automatically locating the centromere in a microscopic image of a human chromosome was presented. Centro mere locating is important for feature extraction and classification of the chromosomes, which is a necessary step towards automatic Karyotyping. The algorithm is based on the calculation and analyzing the vertical and horizontal projection vectors of the binary image of the chromosome. The binary image is obtained by applying a threshold on the input image after histogram modification and analyzing. Chromosome matching can also be down by finding area and lengths and areas of p-arm and q-arm and chromosome central pixel values.
The automatic karyotyping systems allow countless advantages such as interactive and graphical environment, faster in accomplishing the exams. Better interpretation of the image and it still makes possible the storage of the information in a database for future analyses.
5. REFERENCES
[1] Ronald J.Stanley and James M. Keller, Data-Driven Homologue Matching For Chromosome Identification, IEEE Transctions on Medical Imaging Vol. 17, No.3, June-1998.(2)
[2] S.R Ghaffari, Cancer Institute, School of Medicine, University of Tehran, Iran Proceedings of the 16th IEEE Symposium on Automatic Locating the Centromere on Human Chromosome Pictures Computer-Based Medical Systems(CBMS™03) 1063- 7125/03 $17.00 © 2003 IEEE
[3] Yu-Ping Wang, Chromosome image enhancement using Multi scale differential operators IEEE Transactions on Medical Imaging vol. 22, NO. 5, May 2003.
[4] Gady Agam Geometric Separation of Partially Overlapped Objects Applied to Automatic Chromosome Classification IEEE Transaction on pattern Analysis And Machine Intelligence, vol 19, No. 11, November 1997
[5] P.Mousavi, S.S.Fels, R.K.Ward and P.M.Landsdrop Classification homologous human chromosomes using mutual information maximization.0-7803-6725-1/01/$10.00 © 2001 IEEE.
[6] NEMA, Digital Imaging and Communications In Medicine (DICOM) Introduction and Overview Global Engineering Documents, 1998.
[7] Sir Stanley Davidson, Davidsonâ„¢s Principles and Practice of Medicine. Seventeenth Edition, published by Edward, Bouchier ,Haslett and Chilvers
[8] http://ph.tn.tudelft.nl/course/F IP/frames/#Heading118
[9] Gonzalez and Woods, Digital Image Processing 2nd Edition, Published by Addison Wesely Publishing company.
[10] http://biology.arizona.edu/humanbio/acti...yping.html
[11] http://faoDOCREP/004/T0117 E/T0117E11.htm
[12]C.H.Z.pantaleao,Development of a computerized system for cytozenetic analysis and classification0-7803-7612- 9/02/$17.00©2002 IEEE
[13]Noise reduction in alphanumeric characters and handwritten signatures using singular value decomposition. Anderson Duarte Meira, Department de Ciência da Computação
