You are on page 1of 4

International Journal of Computer Information Systems, Vol. 4, No. 2, 2012

Using Local Histogram and FCM Technique for Object Extraction in Semen Sample Microscopic Images, Based on Mathematical Morphology

Mohammad Hossein Bamneshin

Computer Eng. Department Bu-Ali Sina University Hamedan, Iran m.bamneshin@basu.ac.ir

Abdolhamid Pilevar

Computer Eng. Department Bu-Ali Sina University Hamedan, Iran pilevar@basu.ac.ir

Asghar Pirzadeh

Dept. of Hematology and Oncology

Ardabil Medical Science University Ardabil, Iran Asghar.pirzadeh@arums.ac.ir

AbstractNothing that an effective cure for infertility happens when we can find a unique solution, a great deal of study has been done in this field and this is a hot research subject for today studies. So we could analyze the men’s seaman and find out about fertility and infertility and from this find a true cure for this, since this will be a non invasive and low risk procedure, it will be greatly welcomed. In analysis of semen sample images taken from microscope in laboratories, that’s important to achieve correct and accurate sperm segmentation from its fluids and surroundings and will be helpful to result in correct decision in infertility. Some problems such as unsuitable light, images with variable quality and sometimes bad quality affect making threshold for converting images, as a first stage of object tracking. In this research, a dynamic method to making threshold value is presented that by considering images with variable quality and sampling terms, determine appropriate dynamic threshold. Final result of this method comparing with experts outcomes, shows reliable answers.

Keywords-(CASA) computer assisted sperm analysis, threshold in digital image, FCM technique, segmentation, sperm identification.

I.

INTRODUCTION

Infertility is a problem that face it some couples after marriage. Considering high percentage of problems with infertility in couple are from the male and finding ways to resolve this will be helpful to the physicians for a better and faster cure for couples. Whatever is done under meaning infertility analysis in laboratories, contain a process that the sperm slide is being analyzed by a specialist under microscope. But, something that are discussed as problems in this field, is cause of an attempt to finding new methods with good answer in facing with these challenges. These problems are:

- being time consuming the process of sperm analysis under microscope. - A lot of errors which majority of them will be caused by human errors such as boredom and insufficient knowledge.

-

grouping

by

different

people

with

different

observations result in different outcomes for same input

sample.

According to above reasons, it seems necessary to find a method for replacing with human-based methods in this field. This method should be includes some features, such as doing computerizing and classification correctly. This research aims to do appropriate operations on different semen sample microscopic images using image processing algorithms and techniques to present a unique and high performance method. The first step in this, is distinguishing the correct sperm and the fluid surrounding by computer. This action is called as segmentation in image processing field. In the process of the segmentation of the image, one of the most important steps is for distinction and pattern recognition, and the success in the segmentation will guarantee the final result. In this paper we have introduced a unique method for segmentation of sperm based on dynamic threshold form the surrounding area and at last, comparing and discussion on the results.

II.

MATERIALS AND METHODS

  • A. Image acquisition techniques

The first step of imaging is the preparation of the sperm; therefore, first of all, we have to describe the preparation. The sperm remains alive out of the human body for 45 minutes and, during this period, it has a movement. The first step for preparation of an image capable is to use its morphological studies, is fixing the sperm. Note that sperms have a cyclic movement in their medium; They are fixed by use of staining methods and, for the purpose of homogenization, they are put in an incubator for a period of 30 minutes, But it has been proved that the staining methods have no significant role for determination of sperm specifications, so , in some research centers, such as Isfahan

February Issue

Page 39 of 62

ISSN 2229 5208

Infertility and Fertility Center, for the purpose of avoidance form quality decrement of the image, they are not stained, Of course, sperms have a three dimensional movement, in some cases when they are deeper, some image might not show the accurate specifications of the sperm. In some frames of imaging the sperm may produce an image that does not represent its accurate morphological specifications but some other ones may represent its accurate specifications. Therefore, in determination of the accurate morphological specifications of the sperm we can’t rely on a unique image. The videos used in this research have been recorded from Fatemiyeh Hospital IVF center in Hamedan and Didevar Pathologic laboratory in Ardabil. The sample preparation technique was almost the same in both laboratories: the sample was first put in an incubator for 30 minutes. Once it was homogeneous, a drop of 20µl of the sample was put on a clean slide. The images were prepared by use of microscopes equipped with camera at ×40 magnification to enable us to study the sperm using uniform samples [7]. To make the videos, two same microscopes and 2 different cameras were used with the following specifications:

1- Same microscope is used in two place: Olympus tri-

ocular invert microscope with a total magnification of ˣ400 (objective lens: ˣ40, eyepieces lens ˣ10) in model CX21.

2- Moticam 2.0 digital video camera designed for microscopes with 3.2MPixel resolution in Ardabil Center. 3- Sony digital video camera capable to be attached to a third eyepiece.

  • B. Software for algorithm implementation

Proposed algorithm was run using MATLAB 7.12.0.635, with Windows 7 operating system on a Core2Duo 2.8GHz personal computer.

  • C. Edge Extraction

1)

Image Enhancement Method:

In the images taken by the microscope there were a lot of excess parts that needs to be eliminate at first, so there is no confusion with the sperms and in turn does not create errors, also we needed to have the image as monotonous as possible for example small particles on the slide, artifacts cause by light source and sampling errors, and similar errors. We also have to note that the liquid is monotonous and other artifacts like blood cells are not present in our samples.

For improved images we used a median filter with windows of 3x3 to create a monotonous image and eliminate small spots, of course this filter will cause the sharp edges to be disappeared however this will not make much changes to the algorithm of the edges.

International Journal of Computer Information Systems, Vol. 4, No. 2, 2012

Followed by the step above for the improved images and elimination of the sharp edges and noises and also filling the empty spots and elimination of rice artifact noise we used the following method. First we applied a Closing algorithm by use of a disc

element [1].

shows Closing,

shows Dilation, Θ shows Erosion

These narrow and long dents will be connected and will fill all the small cavities in the images and also fill all the small holes that are in the form of artifacts notices.

Then we used a special filter for the resulted images. And finally, applied opening algorithm with the same structuring element on the images with the preliminary size (the size of image are changed as a result of closing).

shows Opening,

shows Dilation, Θ shows Erosion

This action will create a flattened curvature, and if there are narrow spots out of the objects, it will be eliminated [4].

2)

Sperm Identification Method:

The detection threshold method was used for determination of object edges. After completion of the said steps, we applied the threshold method on border illumination levels histogram; a simple and usual way for separation the objects from the background is the selection of a known threshold to separate two phases from each other. The semen image should be divided into a foreground image and background image. The foreground image is the set of all particles in the image and the background image contains the semen plasma. In the image used in this research and in usual form of semen analysis, most of the cells have a lower intensity level than the background. Therefore, the image contains some particles with a homogeneous intensity and a background with a different intensity level. Such an image can be divided into parts through simple threshold:

(3)

That in above relation, input(x,y) is the resulted image from previous stages and image(x,y) is the result image after applying threshold.

However the semen sample images usually have a bimodal histogram, but we should note that determining suitable single threshold for these types of images is important. For explain the reason of importance, we should know that the main activity of threshold value act in boundaries of object wherever background and cells are separating from each other. Considering that the cells have a lower intensity level than the

February Issue

Page 40 of 62

ISSN 2229 5208

background, by using a threshold value near background pixel intensity result in separated cells bigger that their actual size and with threshold value near cells pixel intensity, size will be smaller. Moreover, there are not exact values for peak of each mode in image histogram. Third problem is variety between image intensities that cause different histograms and needs for different threshold values. In this research for resolving these problems, a threshold method using maximum difference with minimum errors between histogram peaks are applied. As said before, this type of image has two major intensity levels. So, it seems existence of two sensible peaks is obvious. The relation below shows the way of computing the proposed threshold (thr):

Thr = min{2C 1 -min , 2C 2 -max} + |2C 1 -2C 2 +max-min|

*

(4)

c 1 and c 2 are the peaks of histogram, max and min are the maximum and minimum intensities of the input image.

background, by using a threshold value near background pixel intensity result in separated cells bigger that

Figure 1: extracted head of special sperm from background using implementation of proposed enhancement and threshold algorithm on input image.

After implementation of the above stages of algorithm, all the background is appeared in a single color (black) and the illumination level of the object is appeared in another color (white). Finally, after using a typical edge detection algorithm such as sobel, an edge formation for the object is appeared.

background, by using a threshold value near background pixel intensity result in separated cells bigger that

Figure 2: result of applying simple edge detection methods on figure 1.

Sometimes, a single threshold is used for all the images [10] that will be involved problems in different light condition sampling and variable images. But because of dynamic threshold determination and image-based operations, above- mentioned algorithm is free from these kinds of problems.

background, by using a threshold value near background pixel intensity result in separated cells bigger that

Figure 3: resulted image of a head of special sperm after segmentation.

International Journal of Computer Information Systems, Vol. 4, No. 2, 2012

D. Recognition of all the sperms in one sample image

By the above threshold method, we can find not only one image but also all the needed images, because of the almost same condition of sperms on image such as the light spectrum and focus, the chosen threshold will be similar for all of them and that means, we will be able to do this for all the sperms on image to be segmented. At first, by clicking on one chosen sperm, algorithm will plot a square with bigger than sperm size around the sperm. Then histogram of the image obtains and parameters of threshold computing will be extracted from it. At the end,

computed threshold imposes on all image and finally all sperms in concerned image are segmented.

III.

RESULTS

In evaluation of the suggested method for sperm head segmentation, we can use two methods: Quantitative and qualitative. from quantitative viewpoint, we can compare results with an expert operator. For this issue, results of applying this method on 37 semen sample compared with the findings from experts in neutral comparison. It was found out that from 37 images a little more that 89% of the result was very much similar under special condition, like bad lighting and low artifact noises. For quantitative evaluation of images, such as most of evaluations, we can note the following factors: total error and precision. Total error means the number of objects or background did not separate in good condition divided by total points should be separate and precision means the number of objects or background separated in good condition divided by total points should be separate. These two factors are based on four amount that are easily computable; false positive (FP), false negative (FN), true positive (TP) and true negative (TN) (first two amount are added for computing total error and two remain are added for determining precision). In image processing, two factors have been applying for evaluation: specificity and sensitivity which are based on four above amounts [9].

Sensitivity defines as:

 

(4)

And Specificity defines as:

 

(5)

February Issue

Page 41 of 62

ISSN 2229 5208

The results shown in bellow table are based on calculating 37 worse images that are selected from 50 random images from 100 semen sample videos.

TABLE I.

THE RESULTS FOR 37 RANDOM IMAGES (PERCENTAGE)

 

Precision

86.11

 

Total error

13.89

Sensitivity

89.74

Specificity

86.36

Also, for 23 Sperm images

with 2

repeat for

each one,

85.7% and with 3 repeat 73.3% of repeatability achieved.

IV.

DISCUSSION AND CONCLUSION

In this research we have taken pictures by microscope for segmentation of sperm. So the instrument we used in this research and the way of our usage and the quality of pictures which we have used as an input data are important processing and the final results. In this project, we use three sets of cameras and microscopes, available in this research centers. For having better result we need to have advanced instrument, so we applied several ways of image enhancement to improve our pictures. As most previous methods steps of our algorithm will

International Journal of Computer Information Systems, Vol. 4, No. 2, 2012

remove sperm tail and middle part of it which was not clear earlier, but show the head as most important part of the sperm is done better than before. For improving results in current situation, recommend to present and to use better image enhancement and segmentation algorithms and better

instruments. And final goal in this research is to reach a

complete algorithm to identify all part of sperm including tail,

middle and head of it separately.

REFERENCES

[1]

Rafael C. Gonzalez, Richard E. Woods Digital Image processing.

[2]

Robert C., Hilborn Chaos and Nonlinear Dynamics Second Edition.

[3]

Metin Akay, Dartmouth College Hanover, NH Nonlinear Biomedical

[4]

Signal Processing. Acosta A. A. and Kruger T F 1996 Human Spermatozoa in Assisted

[5]

Reproduction 2 nd edn (London: Parthenon). WHO 1999 Laboratory Manual for the Examination of Human Semen

[6]

and Sperm-Cervical Mucus Interaction. Teifoory N., Moradi M.H. and Nafisi V.R., 2002 “A new method for

[7]

sperm segmentation in microscopic image” Proc. 11 th Iranian Biomedical Engineering Conference Amirkabir University of Technology. Nafisi V.R., Moradi M.H. and Nasr-Esfahani, 2005, “A template

[8]

matching algorithm for sperm tracking and classification.”, Physiological Measurement. Sedighi S., Moradi M.H. and Nafisi V.R., 2004, “Compare several

[9]

methods for sperm segmentation in microscopic image” Master project, Islamic of Azad University of Technology. Anke Meyer-Base, Pattern Recognition for Medical Imaging”, Department of Electrical and Computer Engineering, 2004, Florida state University Tallahassee.

February Issue

Page 42 of 62

ISSN 2229 5208