You are on page 1of 5

International Journal of

Science and Engineering Investigations

vol. 1, issue 10, November 2012 ISSN: 2251-8843

Mammographic Classification Based on Fuzzy Method and Histogram Information


Iman Rahimi 1, Hamed Rahimi 2, Sara Soheili 3
1

Dept. of Mechanical and Manufacturing Eng., Faculty of Eng., Univ. of Putra Malaysia (UPM), Serdang, Selangor, Malaysia 2 Faculty of Science, Qom Branch, Islamic Azad University, Qom, Iran 3 Dept. of Medical Microbiology & Parasitology, Faculty of Medicine & Health Sciences, Univ. of Putra Malaysia (UPM), Serdang, Selangor, Malaysia
(1imanrahimi64@yahoo.com, 2pcdp08@yahoo.com, 3sara_soheilii@yahoo.com)

Abstract- Breast cancer risk is identified by many signs especially mammographic density. Histogram information research lead to some quantitative forecasting approaches, many of researchers interpret the consequences arising from grey level information ineffective for discerning between complicated density classes. A technique is using fuzzy logic to determine the threshold of gray images and then to get the ROI (Region of Interest) of them and use of Histogram. Keywords- Automatic Threshold Detection, Breast Cancer, Fuzzy Logic, Mamographic Density, Pattern Recognition.

mammogram that it will be used only from useful area of image and later can be seen that it increases speed of classifier. In Figure 2 histograms of images from A to C and D to F can be seen.

I.

INTRODUCTION

Density of the mammogram is known to be as an important sign of breast cancer. Some examples of different mammographic densities can be seen in Fig. 1. Mammographic images are 2D projections of the x-ray attenuation properties of the 3D breast tissue along the path of the x-rays. The connective and ephiteal tissues attenuate more than fatty tissues. In fact, This phenomenon can be interpreted that brighter areas on mammographic films represent glandular tissues, whereas the dark areas represent fatty tissues. From an image analysis point of view, it can be seen that mammographic densities correspond to image intensities. This leads to the modeling based on histogram information. Since many studies for classification the type of mammogram density according to Histogram information is done. Since Wolfe has introduced the mammographic risk assessment [1], automatic breast parenchymal pattern classification methods have been investigated. for example: Byng et al. proposed an interactive thresholding technique [2], Karssemeijer proposed to use the skewness, rather than the standard deviation, of histograms of local breast areas [3], Masek uses the average of original image Histograms as a features vector for classifier [4] Study by Zhou [5] showed that for each class Histogram patterns density is different, to obtain Histogram information first by using fuzzy method a threshold value for each image is obtained [6], then using this ROI (Region Of Interest) threshold value of image and ROI Histogram of image obtained using ROI of image because of additional and waste information in

Fig. 1 Example mammograms: (a,d) Fatty, (b,e) Gandular, and (c,f) Dense.

86

In practice radiologists are evaluated that the density value not solely based on light and examine tissue pattern and tissue distribution. So background information plays a role in differing density. The main goal of this study using multi resolution Histogram technique for classifying types of mammogram density is also proof that multi resolution Histogram is more effective than Single Resolution Histogram. [7] many studies has been done to obtain the threshold value [8] main goal to obtain the threshold value is classification pixels of given image in two parts that first is an object and a second is the background, if one of This section includes the amounts that are smaller than or equal to the threshold value the second part is include larger value of the threshold. To obtain the threshold value, entropy of image should be used, there is a lot of entropy for this action but we use sure entropy [9], which has the following:

Here, y is the independent variable, a and c are parameters determining the shape of the above two membership functions as shown in Fig. 3. The membership function of the fuzzy set dark in W can be interpreted as the membership of black pixel group, while the membership function of the fuzzy set bright denotes the membership of the white pixel group.

is a positive threshold value and based on many studies have been done set it to 2[9]. Also Shannon entropy can be used as follows [11,10] :

(a)

And even Charvat entropy can also be used, which in [12]:

Note that generally set n equal to 0.1.

II.

ALGORITHM

(b)
Fig. 2. Example histograms for the MIAS density classes. The three histograms in (a) represent the fatty (continuous line), glandular (dashed line), and dense (dotted line) mammographic examples from Fig. 1.a-c and show good separation with respect to greylevel. The three histograms in (b) represent three other mammograms (using the same key for fatty, glandular and dense mammograms) from the MIAS database which are displayed in Fig. 1 d-f for which it might be difficult to separate them just on greylevel information.

A level of gray is composed of J level color from r0 to rj-1 and call the Histogram of image hdthat d = 0, ...,j-1. Show the image with (W, P) that W = {r0, r1, ..., rj-1 } and P is the event probability of a color level and Pr {rk}= hk. Also B (rd) is a membership function of fuzzy set, B W is the grade properties of darkness and lightness of image, and also for a level of color rd in gray image we have a fuzzy set [13]: B=B(r1)/r1+B(r2)/r2++B(rd)/rd. To obtain the threshold value Fuzzy Method divided pixels to bright and dark area and W, whose membership functions are defined as below:

If new computed H is greater than Hmax, we replace Hmax with new computed H. In the same way, it is replaced a opt, and copt, with new computed a and c, respectively. End for a End for c Threshold value is: t = (aopt + copt)/2. Create desired mask with above threshold value Figure7 shows membership function for image of Figure 5. Compute the histograms {h0 , h1 , h2, h3, h4}, where hi is a row vector related to the histogram of resolution i and i =0 is the highest resolution.

International Journal of Science and Engineering Investigations, Volume 1, Issue 10, November 2012 ISSN: 2251-8843 www.IJSEI.com

87 Paper ID: 11012-19

Fig 3.Membership Function of Dark And Bright Area

Fig 6. Created Mask via Fuzzy method For image of figure 5

Fig 4. Algorithm Flowchart

Fig 7.membership Function Of Image figure 5

Compute difference histogram between consecutive levels. The difference histogram has length of 1024 .see figure 10. Sub-sample difference histograms and renormalize. We used a sub-sampling factor of 22/3, the sample size due to reduced feature vector and thus speed up classifier see figure 11.

III.
Fig 5. Gaussian pyramid of mammogram

DATA

Normalize histograms with Lnorm. This is similar to normalize the histograms with respect to the breast area. Figure 8 is normalize histogram of figure 5. Compute cumulative histograms. see figure 9.

We evaluated the proposed technique on 321 mammogram images of the Mammographic Image Analysis Society (MIAS) database and classified them based on the provided three class categories (Fatty, Glandular, Dense)[15,16]. It should be noted that the database has equal proportions for these three classes.

International Journal of Science and Engineering Investigations, Volume 1, Issue 10, November 2012 ISSN: 2251-8843 www.IJSEI.com

88 Paper ID: 11012-19

IV.

EVALUATION

For classifying to mention three Categories, data of the last step of algorithm give to SVM as an input to classify them to three categories.

fig 8.Normalize Histograms for image of figure 5(highest Histogram has the highest resolution) .

Fig 11. Sub-sampled Difference Histogram

The MIAS database consists of pairs of mammograms; hence, to minimize bias, we treated the left and right mammograms independently. Thus, we trained the classifier on all the left (right) mammograms according to a leave-oneimage-out methodology. And then we combine the results of the left and right mammograms in the dataset. The results are presented in the form of confusion matrices.

V.
Fig 9. Cumulative histograms of figure 8.

CONCLUTION

Results for the MIAS database of pictures without the use of Fuzzy ROI and with Gaussian kernel of SVM is shown in Table 1.

Table 1

Classified Base on SVM

Fatty Glandular Dense

MIAS Database Classification Fatty Glandular Dense 93% 5% 2% 15% 60% 25% 3% 17% 80%

However, the result for the MIAS database of images without the use of Fuzzy ROI and with polynomial kernel of SVM is shown in Table 2 is shown.
fig 10.Diferece histograms of figure 9

K(x,y)=(1+xTy)p

International Journal of Science and Engineering Investigations, Volume 1, Issue 10, November 2012 ISSN: 2251-8843 www.IJSEI.com

89 Paper ID: 11012-19

Table 2

REFFERENCES
[1] J.N. Wolfe, Risk for breast cancer development determined bymammographic parenchymal pattern, Cancer, vol. 37, pp. 2486 2492, 1976. J.W. Byng, N.F. Boyd, E. Fishell, R.A. Jong, and M.J. Yaffe, Thequantitative analysis of mammographic densities, Physics inMedicine and Biology, vol. 39, pp. 1629-1638, 1994. N. Karssemeijer,, Automated classification of parenchymal patternsin mammograms, Physics in Medicine and Biology, vol. 43, pp. 365 -378, 1998. M. Masek, S.M. Kwok, C.J.S. deSilva, and Y. Attikiouzel,Classification of Mammographic Density Using Histogram DistanceMeasures, in CD-ROM Proceedings of the World Congress onMedical Physics and Biomedical Engineering, August, 2003, pp. 1. C. Zhou, H.P. Chan, N. Petrick, M.A. Helvie, M. Goodsitt, B.Sahiner, and L.M. Hadjiiski, Computerized image analysis:Estimation of breast density on mammograms, Medical Physics,vol. 28, no. 6, pp. 10561069, June, 2001. EnginAvci, DeryaAvci, " An expert system based on fuzzy entropy for automaticthreshold selection in image processing" , Expert Systems with Applications 36 (2009) 30773085 Izzati Muhimmah, ReyerZwiggelaar," Mammographic Density Classification using MultiresolutionHistogram Information" , IEEE 2007 Chang, C. I., Chen, K., Wang, J., &Althouse, M. L. G. (1994). A relativeentropy-based approach to image thresholding. Pattern Recognition,27(9), 12751289. Turkoglu, I. (2002). An intelligent pattern recognition for nonstationarysignals based on the timefrequency entropies. PhD Thesis, FiratUniversity Graduate School of Natural and Applied Sciences, Elazig, pp. 4047. Bezdek, J. C. (1987). Partition structures: A tutorial, analysis of fuzzyinformation (Vol. III). Boca Raton: CRC Press. Cheng, H. D., Chen, J. R., & Li, J. (1998). Threshold selection based on fuzzy c-partition entropy approach. Pattern Recognition, 31(7), 857 870. Karmeshu, J. (2003). Entropy measures, maximum entropy principle and emerging applications. New York: Springer-Verlag Berlin Heidelberg. Zadeh, L. A. (1973). Outline of a new approach to the analysis of complex systems and decision processes. IEEE Transactions on Systems Man and Cybernetics, SMC-3(1), 2844. Karmeshu, J. (2003). Entropy measures, maximum entropy principle andemerging applications. New York: Springer-Verlag Berlin Heidelberg. Kittler, J., & Illingworth, J. (1986). Minimum error thresholding. PatternRecognition, 19(1), 4147. E. Hadjidemetriou, M.D. Grossberg, and S.K. Nayar, MultiresolutionHistograms and Their Use for Recognition, IEEE Transactions onPattern Analysis and Machine Intelligence, vol. 26, no. 7, pp. 831-847, July, 2004. J. Suckling, J. Parker, D. Dance, S. Astley, I. Hutt, C. Boggis, I. Ricketts, E. Stamatakis, N. Cerneaz, S. Kok, P. Taylor, D. Betal, and J. Savage, The Mammographic Image Analysis Society Digital Mammogram Database., ExcerptaMedica. International Congress Series, vol. 1069, pp. 375-378, 1994.

Classified Base on SVM

Fatty Glandular Dense

MIAS Database Classification Fatty Glandular Dense 97% 3% 0% 5% 65% 35% 12% 18% 70%

[2]

[3]

But when we use the fuzzy method to obtain the threshold value and then ROI of image we get better results, Results for MIAS database images using fuzzy ROI and with Gaussian kernel of SVM is Shown in table 3.

[4]

[5] Table 3

Classified Base on SVM

Fatty Glandular Dense

MIAS Database Classification Fatty Glandular Dense 98% 2% 0% 7% 68% 25% 10% 20% 70%

[6]

[7]

[8]

And final result for MIAS database images using fuzzy ROI and with polynomial kernel of SVM is shown in table 4.

[9]

Table 4

[10] [11]

Classified Base on SVM

Fatty Glandular Dense

MIAS Database Classification Fatty Glandular Dense 98% 2% 0% 10% 70% 20% 8% 15% 77%

[12]

[13]

VI.

SUMMARY

[14]

We review in this paper classifying mammography images into three class of fatty, Glandular and dense, also for improve Results we use ROI of image with Fuzzy method, testing Kernel of SVM and using multi-resolution Histogram that each of them improve results. The new work in this paper was using fuzzy method to determine the threshold value and then get a ROI of useful area to get Histogram of images.

[15]

[16]

International Journal of Science and Engineering Investigations, Volume 1, Issue 10, November 2012 ISSN: 2251-8843 www.IJSEI.com

90 Paper ID: 11012-19

You might also like