Images Enhancement ofO-band Chromosome Using histogram ... · adaptive and iterative histogram matching (AIHM) algorithm for chromosome contrast enhancement especially in banding

Images Enhancement ofO-band Chromosome Using

histogram equalization, OTSU thresholding,

morphological dilation and flood fill techniques

Pichet Wayalun

Information Technology Division Mahasarakham University

Mahasarakham, Thailand 44150 Email: [email protected]

Abstract- Human chromosomes contain important information

for cytogenetic analysis. Cytogenetic compares their patient's

chromosome images against the prototype human chromosome

band patterns. Chromosome images were acquired by

microscopic imaging of metaphase or prophase cells on specimen

slides. Digitized chromosome images usually are suffered from

poor image quality, lack of contrast and hole in images.

Therefore, the images must be enhanced. This paper presented

an enhancement algorithm for chromosome images based on

histogram equalization (HE) OTSU flood-full and dilations.

Firstly, the histogram equalization is used to improve the

contrast of the image. Then the OTSU threshold segment, flood­

fill fill hole in image and dilation are experimented to validate the

effect of this algorithm, the canny edge detection and oriented

bounding box were used to detect the edge and segment

chromosomes from background. A success rate of the proposed

method can be improved to 97.07% of the enhancement in the

chromosome images.

Keywords- chromosome; chromosome karyotyping; mathematical morphology; enhancement chromosome;

I. INTRODUCTION

Chromosome analysis is an important task in clinical and research cytogeneticists. The patterns of chromosome bands provide the essential visual information for cytogeneticists to examine the genomic composition of cells, and for feature measurements in automated karyotyping instruments. There for, the images are hardly usable for visualization and karyotyping purposes. Compute processing and enhancement of chromosome images can largely remedy this problem. Currently, available systems employ conventional convolution filtering and contrast stretching, to compensate for the image quality deficiencies [I]. Wang et al. [2] apply a class of differential wavelets chromosome images enhancement. The differential wavelet representation provides high-frequency edge information along horizontal, vertical, and diagonal directions, used in the design of an image-enhancement algorithm. The data set for test is 342 G-banded cells containing 15,136 chromosomes. Wenzhong [3] applies some

163

Phatthanaphong Chomphuwiset, Natthariya Laopracha and Phatthanaphon Wanchanthuek

Computer Science Division Mahasarakham University

Mahasarakham, Thailand 44150 Email: {phatthanaphon.c & natthariya &

pwanch }@gmail.com

mathematical morphologies to enhance the chromosome images. The method consists of the top-hat transform and bot­hat transform, which arms to improve the contrast of the images. Then the iterative threshold segmentation, i.e. closing operation and opening operation, are performed on the result of this algorithm. The standard canny edge detection operator is exploited to detect the edge of chromosomes. Wang et al. [4] use some oriented wavelets to derive from isotropic Laplacian-like filters to enhancement chromosome images. Wu and Castleman [1] applied cubic-spline wavelet transforms and multiscale-solution image analysis techniques to enhancement chromosome images. PAL et al. [5] uses some minimization of compactness and fuzziness to enhancement chromosome images. Wu et al. [6] applied some wavelet-based to enhancement chromosome images .This technique employs a wavelet transform derived from a special class of differential operators to generate a multi-scale representation of images. A multi-scale point-wise product (MMP) is used to characterize the correlation of the image features in the scale-space. The data set test has 342 G-banded cells containing 15,736 chromosomes. Choi et al. [7] suggest a fuzzy logic approach to enhance the image. In this approach, the selection criteria constitute the antecedent clauses of the fuzzy rule, and the corresponding filters constitute the consequent clauses of the fuzzy rules. Ehsani et al. [8] used to adaptive and iterative histogram matching (AIHM) algorithm for chromosome contrast enhancement especially in banding patterns. In the evaluation of the proposed method, a set of human single chromosomes were tested, which consists of 30 chromosomes of different.

However, a number of research paper have not raised such the issue and attempted to solve it. The main objective of chromosome image enhancement is to improve the visibility of details while suppressing noises or to reduce unrelated artifacts such as holes in images. Among various techniques for global enhancement and fill hole, histogram equalization and flood fill are widely used method [9-11].

Chromosome images were acquired by microscopic imaging of metaphase or prophase cells on specimen slides. Digitized chromosome images usually are suffered from poor image quality, lack of contrast and hole in images. Therefore, the images must be enhanced.

This paper makes uses of Histogram equalization (HE) flood-full and dilations for images enhancement applications by exploiting the geometric correlation of image features in the domain. The resulting enhancement scheme is particularly applied to chromosome image enhancement. The remainder of the paper is organized as follows. Basic definitions are given in section 2. Section 3 presents a novel method for enhancement the chromosome images. The performance of the method is analysed in section 4. Finally, section 5 gives conclusion.

II. THE ENHANCEMENT ALGORITHM

In our algorithm, we apply histogram equalization, flood­full and dilation to enhance the chromosome images. Firstly, the histogram equalization is used to improve the contrast of images. Then, the OTSU threshold [12] segments images from background, flood-fill [13] fills hole in images and dilates images by using a specific structuring element. In our enhancement algorithms follow the same core steps as show in Figure l.

original image

I

enhancement lmage

Figure 1. Enhancement algorithm schema

A. Histogram equalization (HE)

Considering a digital image with gray levels in the range [0, L -1], Probability Distribution Function of the image can be computed as equation (1):

nk perk ) = - k = O, . . . , L - I

N (1)

Where kr is the kth gray level and nk is the number of pixels in the image having gray level rk. Cumulative Distribution Function (CDF) can also be computed as follow:

;�k C(rk) = LP(lf) (2)

;�O

164

k = O, ... ,L -1, 0 S C(rk) S 1 Histogram Equalization (HE) appropriates gray level Sk to

gray level rk of the input image using equation (2). We therefore have:

(3)

Gray level Sk be computed in usual histogram equalization method as:

(4)

Equation (4) means that distance between Sk and Sk-l has direct relation with PDF of the input image at gray level rk.

Undesirable effects of the usual histogram equalization method (HE) are resulted from equation (2) because the quantization operation and summarizing properties of this equation [14].

B. Otsu Method

Otsu method proposed by Lin [12] as follows. Assuming an image is represented in L gray levels [0,1, . . . L-l]. The number of pixels at level i is denoted by ni , and the total number of pixels is denoted by N =nj+n2+n3 ... +nL. The probability of gray level i is denoted as

L-I p; = n; / N,p; � 0, LP; =1 (5)

o

In the bi-level threshold method, the pixels of image are divided into two classes C1 with gray levels [0,1, . . . t] and C2 with gray levels [t+l, . . . L- l] by the threshold t . The gray level probability distributions for the two classes are

I WI = Pr(CI) = L P;

;�O

L-I w2 = Pr(C2 ) = L P;

;�I+I

The means of class C1 and C2 are

t U1 = L ip; /W1

i�O

1.-1 U2 = L iPi /W2

i�t+l

The total mean of gray levels is denoted by Ur

The class variances are

(6)

(7)

(8)

(9)

(10)

t a} = I (i -u1)2 Pi / Wi

i�O 1"-1

(J"� = I (i -u2)2 Pi / W2 i�t+l

The within -class variance is

The between-class variance is

(J"� = Wj (uj -Ur)2 + w2 (u2 -Ur)2 The total variance of gray levels is

(11)

(12)

(13)

(14)

( 15)

Otsu method can be extended to multilevel threshold method. Assuming that there are M-I thresholds [tIh ... tM-I] that divide the pixels in the image to M classes {CI • C2, ... CM }.

Where

{tl+t2°o. tM_d=arg { max {(J"�(tl,t2 .. .tM-l )}} O:;t:;L-l

W = ]

U = ]

;=Ij_1 +1

i=lj_l+1

IJ (J"; = I (i -ul)2 Pi / W]

i�li+l

M

(J"� = "w(u -ur)2 L. ] ] j�l

C. Flood Fill

(16)

(17)

(18)

(19)

(20)

(21)

Flood-fill [15] is a technique that determines the area connected to a given node in a multi-dimensional array. The technique looks for all nodes in the array connected to the start node (called seed point) by a path of the target color, and changes them to the replacement color.

165

For example, the area is surrounded by the black pixels in Figure 2 (a). The red pixel in Figure 2 (b) represents the seed point that is already changed to the replacement color (red). Figure 2 (c) shows the final result after flood-fill.

(a) (b)

(c)

Figure 2. Example of queue-linear flood fill algorithm

D. Dilation

Mathematical is a methodology for extracting shape and size information from an image. It involves configuration of a set of nonlinear operators that act on images by using structuring elements. The one basic morphological operator is dilation, from which many operations can be derived. For grey-scale image f and structuring element B, dilation is defined as follows:

dilation: 6B(f)(x,y)

=

{f(X-S, y-t)+B

(S,t)1 }

max (x-s),( y-t) E D

f;(s,t) E DB

(22)

where for imagefand structuring element B, (x, y) and (s, t)

are the respective co-ordinate sets and Dfand DB are the respective domains [16].

E. Image Enhancement

In order to improve the image contrast, it must be segmented by exact threshold. In this method, we used the Otsu threshold algorithm to segment the image. First, the original images add the result of histogram equalization. Then the result is segmented by the Otsu threshold algorithm. Figure 3 shows the result of the histogram equalization and segment by the Otsu threshold algorithm.

(a)

(b) (c)

Figure 3. Result of the method (a) Original image, (b) Result of HE, (c) Result of OTSU

After the segmentation, the holes in chromosome images are acquired. Then the flood-fill algorithm is applied to fill hole with segmented and dilated chromosome area image by dilation. Figure 4 shows the result of the flood-fill algorithm and dilated chromosome area image by dilation operation.

(a) (b)

Figure 4. Result of the enhancement (a) Result of flood-fill, (b) Result of dilation operation

It can be seen from the Figure4 that the contrast of the image is improved effectively. This will facilitate the next processing.

III. EXPERIMENTAL RESULTS

A. Data set

The proposed system has been tested with a dataset consisting of 917 chromosomes from 20 cells provided by the

166

Center of Medical Genetics Research Rajanukul Institutes of Health. This database consists of 20 cells, 16 males and 4 females. These 20 cells consist of 13 cells with 46 chromosomes, 5 cells with 45 chromosomes and 2 cells with 47 chromosomes. There are 4 cells with 46 chromosomes having strange constellations. The set of cells with one extra chromosome is composed of 4 Down, 2 Edwards and P1itau syndromes, one trisomy 15, three trisomy 16 and two trisomy 10.

B. Performance Study

Experiments are performed on a PC with Intel (R) Core (TM) 3.30 GHz CPU and 2 G main memory, running on Windows 7 .The program are implemented using Opencv 2.1 library in Microsoft visual C++.

To validate the effect of this algorithm, the standard canny edge detection [17] operator and oriented bounding boxes (OBBs) [IS] are used to segment chromosomes from background, and the result can be counted and confirmed by an expert.

The performance of the chromosome segmentation step is evaluated using the overall and chromosomes segmentation accuracy [19], defined as

A chromosome segmentation

X I 00 ccuracy = -

overall_segmentation (23)

The proposed method was compared with three traditional methods; the contrast limited adaptive histogram equalization (CLASE), contrast stretching (CS) and histogram equalization (HE) algorithm. 20 chromosome spread images are tested in our experiment. Figure5-S shows the results comparing to the proposed method with the three methods on single chromosome.

D � �

(a) (b) (c) (d) (e)

Figure 5. Result of the enhancement (a) Original image (b) Result of CLASE (c) OTSU threshold (d) Canny edge detection (e) Result of OBB

Figure 6.

(a) (b) ( c) (e)

Result of the enhancement (a) Original image (b) Result of CS ( c) Canny edge detection (d) Result of OBB

6 I (a) (b) (c) (d) (e)

Figure 7. Result of the enhancement (a) Original image (b) Result of HE (c) Canny edge detection (d) Result of OBB

� (a) (b) (c) ( d) (e)

Figure 8. Result of the enhancement (a) Original image (b) Result of propose method (c) Canny edge detection (d) Result of OBB

Figure 9 shows that the proposed method is the highest performance, and the images data have effect to other method such as contrast and none stable of chromosome distribute.

120

100

f 80

60

if. 40

20

JI( 1\ � �I\' 1\ I�

1 2 3 4 S 6 7 8 91011111314151617181920

chromosome images

�CLASE

...... cs

...... HE

�Proposed method

Figure 9. The result of CLASE, CS and HE and our proposed method

The chromosome image number 19 having high accuracy because it has height contrast and stable of chromosome distribute as show in Figure 10. The image no. 16 is low accuracy because it has low contrast and none stable of chromosome distribute as show in Figurell.

.,

Figure 10. The original image no. 19

167

Figure II. The original image no. 16

The average of accuracy chromosome segmentation values are shown in table I.

TABLE L THE COMPARISONS OF ACCURACY FOR SEGMENTATION

Method Accuracy (%)

CLASE 74.89

CS 93.2

HE 93.32

Proposed method 97.07

From the table I, the proposed method achieves the highest accuracy compared with the other three algorithms. Furthermore, the visual effects of proposed method enhancement are better than the other three algorithms, as shown in Figure 12. This implies that the proposed method is better than the other three algorithms.

( c) (d)

Figure 12. The result of the enhancement (a) result of CLASE (c) result of CS (d) result of HE (e) result of proposed method

IV. CONCLUSIONS

This paper presented an enhancement algorithm for chromosome images based on histogram equalization (HE) flood-full and dilations. To validate the effect of this algorithm, the stated canny edge detection operator and oriented bounding boxes (OBBs) are used to segment chromosomes from background. In order, the result can be counted and confirmed by experts. The proposed method was compared with three traditional approaches, i.e. contrast limited adaptive histogram equalization (CLASE), contrast stretching (CS) and histogram equalization (HE) algorithm. 20 chromosomes of spread images are used to evaluate the proposed method. The results indicate that the proposed method gives 97.07% of accuracy, which is the highest accuracy compared with the other three algorithms.

ACKNOWLEDGMENT

This research was supported by the Center of Medical Genetics Research Rajanukul Institutes of Health. We are also grateful to the reviewers for fruitful comments. This research is funded by Mahasarakham University.

REFERENCES

[I] Q. WU and K. R. Castleman, "Wavelet-Based Enhancement of Human Chromosome Images," in Proceedings of the 20th Annual International

Conference of the IEEE Engineering in Medicine and Biology Society,

1998. [2] Y. P. Wang, et aI., "Image Enhancement Using Multiscale Differential

Operators," in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '01), 2001, pp. 1853-1856.

[3] Y. Wenzhong, "Mathematical Morphology Based Enhancement for Chromosome Images," in ICBBE 2009 3rd International Conference on

Bioinformatics and Biomedical Engineering, 2009 pp. 1-3. [4] Y. P. Wang, et al., "Image enhancement using multi scale oriented

wavelets," in Proceedings 2001 International Conference on Image

Processing, 2001 pp. 610 - 613.

168

[5] S. K. PAL and A. ROSENFELD, "Image enhancement and thresholding by optimization of fuzzy compactness," Pattern Recognition, voL 7, pp. 77-86, February 1988.

[6] Q. WU. et al., "The Effect of Image Enhancement on Biomedical Pattern Recognition.," in Proceedings of the Second Joint EMBSBMES Conference Houston, TX, USA, 2002.

[7] Y. Choi and R. Krishnapuram, "A Fuzzy-Rule-Based Image Enhancement Method for Medical Applications," in Proceedings of the Eighth IEEE Symposium on Computer-Based Medical Systems, 1995 pp. 75 - 80.

[8] S. P. Ehsani, et aI., "Chromosome Image Contrast Enhancement Using Adaptive, Iterative Histogram Matching," in 2011 7th Iranian Machine

Vision and Image Processing (MVIP) 2011. [9] .lames D. Fahnestock and R. A. Schowengerdt, "Spatially variant

contrast enhancement using local rang modification," voL 22, pp. 378-381, 1983.

[10] Rafael C. Gonzalez and R. E. Woods. (1992). Digital Image

Processing.

[11] M. A. Sid-Ahmed. (1995). Image Processing: Theory, Algorithms, and

Archi-tectures. [12] D. Lin and .I. Yu, "Otsu method and K-means," in 2009 Ninth

International Conference on Hybrid Intelligent Systems, 2009, pp. 344-349.

[13] Rafael C. Gonzalez and R. E. Woods. (2002). Digital Image

Processing. [14] H. Yeganeh, et aI., "A Novel Approach for Contrast Enhancement

Based on Histogram Equalization," in Proceedings of the International

Conference on Computer and Communication Engineering, Kuala Lumpur, Malaysia, 2008.

[15] S L. Hsieh, et aI., "Using margin information to detect regions of interest in images," in The IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2011, pp. 3392-3396.

[16] Y. Kimori, "Mathematical morphology-based approach to the enhancement of morphological features in medical images," Journal of Clinical Bioinformatics, vol. 1,20 II.

[17] I Canny, "A computational approach to edge detection," IEEE

Transactions on Pattern Analysis and Machine Interlligence. vol. PAMI-8, pp. 679 - 698 1986.

[18] H. Songhua and Y. Lizhen, "Optimization of Collision Detection Algorithm based on OBB," in International Conference on Measuring Technology and Mechatronics Automation, 2010, pp. pp. 853 - 855.

[19] P S. Karvelis, et aI., "A Multichannel Watershed-Based Segmentation Method for Multispectral Chromosome Classification," IEEE TRANSACTIONS ON MEDICAL IMAGING. vol. 27, pp. 697-708, MA Y 2008 2008