WBC Image Segmentation and. Classification Using RVM

Applied Mathematical Sciences, Vol. 8, 2014, no. 45, 2227 - 2237 HIKARI Ltd, www.m-hikari.com http://dx.doi.org/10.12988/ams.2014.43191 WBC Image Seg...
Author: Morgan Stafford
1 downloads 2 Views 196KB Size
Applied Mathematical Sciences, Vol. 8, 2014, no. 45, 2227 - 2237 HIKARI Ltd, www.m-hikari.com http://dx.doi.org/10.12988/ams.2014.43191

WBC Image Segmentation and Classification Using RVM S. Ravikumar Department of Computer Applications Bannari Amman Institute of Technology Sathyamangalam, Erode(DT), Tamil Nadu, India

A. Shanmugam Principal, Bannari Amman Institute of Technology Sathyamangalam, Erode(DT), Tamil Nadu,India Copyright © 2014 S. Ravikumar and A. Shanmugam. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract Medical Image Segmentation becomes vital process for its proper detection and diagnosis of diseases. In which accurate White Blood Cells segmentation becomes important issue because differential counting, plays a major role in the determination of diseases and based on the treatment is followed for the patients. The Standard Modified Fuzzy Possibilistic C Means are used for segmentation. This paper gives a novel technique for WBC detection based on Relevance Vector Machine (RVM). This proposed method effectively works for WBC detection, and effectively reduces the computational time and preserve the images. In this to improve the efficiency by using three feature vectors it is based on area, length, perimeter etc., and this paper gives the testing efficiency of 91%, compared to existing method this proposed method of RVM gives best result.

Keywords: White Blood Cells (WBCs),Red Blood Cells (RBCs), Modified Fuzzy Possibilistic C Means (MFPCM), Relevance Vector Machine (RVM), Feature Selection

2228

S. Ravikumar and A. Shanmugam

1. Introduction Medical Field is a very important field which has grown tremendously in recent years, with the technical progress in medical field, there is a need for faster and more accurate analysis tool which is essential (e.g. x-ray machines, complete blood count machines…etc). These automated medical tools are necessary for diagnosing patients. They are essential for supporting doctors in accurately providing future prognoses of the conditions and how to cure them [1]. Due to increase in diseases today [1]. There is a need for more medical tools to help doctors to diagnoses fast and accurately. Most of the main laboratories requirements are automated today, and smart systems are used for bone marrow analysis and for differential count of blood components (e.g. to count the number of red and white blood cells, platelets etc..).Red Blood Cells (RBCs), White Blood Cells (WBCs) and blood platelets are the three different types of cells in a human body. In this three types WBC is used for the automated detection. This type of automated detection is one of the chief role to diagnosis the diseases. Numbers of diseases are automated detected by WBC such as lymphoblastic Leukaemia. They are neutrophils, basophils, eosinophils, lymphocytes and monocytes. These five types are categorized into two different major types based on the presence or absence of granules in the cell body. Then another three neutrophils, basophils and eosinophils presence of granules these three are affected by different stains [2].

RBC Platelets WBC

Fig. 1.1 Blood smear image The figure 1.1 shows the real blood smear composition. This is a sample image of a blood smear as seen by the medical microscope that demonstrates typical blood composition, including RBC, WBC and Platelets. A manual count will also give information about other cells that are not normally present in peripheral blood but might be released in certain disease [4]. In particular, the identification and differential count of blood’s cell is a time consuming and repetitive task that can be influenced by operator’s accuracy and tiredness [3]. In

WBC image segmentation and classification using RVM

2229

an effort to overcome the tedious and time-consuming task of human experts in counting white blood cells in bone marrow or peripheral blood, many automated techniques have been proposed[5][6].This is the most difficult part of automatic cell analysis, because there is considerable uncertainty in the microscopic images [7,8 and 9]. In [10] also mention the segmentation accuracy, as the first step in image analysis system which has a very high impact on the whole system efficiency.

2. Literature survey In[11] provides the very easy, fast and inexpensive segmentation process through thresholding. With the help of thresholding methods differentiate the grey level in the image from the background level and also the images are do not touch each other, it is one of the important role of the segmentation by thresholding. Another image segmentation method for white blood cells was implemented in [12], but this also based upon thresholding but that thresholding is depend upon multiple gray level thresholding. Filters are implemented in this method. Low pass filters are used in this method with window size of 5 x 5. In [13] performed comparisons between nine image segmentation which is gray level thresholding pattern matching, morphological operators, filtering operators, gradient-in method, edge detection operators, RGB color thresholding, color matching and HSL (hue, saturation, lightness) and color thresholding techniques on RBC and concluded that there is no single method can be considered good for RBC segmentation [23].These are two difficult issues in image segmentation where common segmentation algorithms cannot solve this problem [10]. Besides that, there are staining and illumination inconsistencies also acts as uncertainty to the image [14]. Sabino and Costa [16] used the Green channel of the RGB model to segment WBC. On the other hand, Westpfalz applied HSI color model to separate WBC from background and de-cluster the clustered WBC [15]. Yang, Foran and Meer [18] improved the algorithm of IGDS to better segment WBC from other BCs presented in the smear image. Sinha and Ramakrishnan [19] used color used HIS equivalent of the WBC image, K-Means clustering followed by EM-algorithm to segment WBC along with the cytoplasm nucleus. In [20] used fuzzy patch labeling to segment WBC from other blood elements.

2230

S. Ravikumar and A. Shanmugam

3. Methodology RGB to HSV Image enhancemen Image t Segmentatio Classificatio n n Figure 3.1 Proposed Methodology The Clustering is the process used in data mining and image processing mostly. They are used to group the objects based on their values or distance etc. For Images the clustering work is to segment the required objects from other objects in an image. This plays a major role in medical image segmentation, satellite image segmentation and many others. In this paper clustering technique is used for WBC segmentation from cell images using various fuzzy based techniques to achieve best clustering technique for WBC segmentation and it is classified by RVM.

3.1. RGB TO HSV A three dimensional representation of the HSV color space is a hexacone, where the central vertical axis represents the Intensity. Hue is defined as an angle in the range [0,2S] relative to the Red axis with red at angle 0, green at 2S/3, blue at 4S/3 and red again at 2S. Saturation is the depth or purity of the color and is measured as a radial distance from the central axis with value between 0 at the center to 1 at the outer surface. For S=0, as one moves higher along the Intensity axis, one goes from Black to White through various shades of gray. When Saturation is near 0, all pixels, even with different Hues look alike and as we increase the Saturation towards 1, they tend to get separated and are visually perceived as the true colors represented by their Hues. (1) In the above equation, we see that for V=0, th(V) =1.0, meaning that all the colors are approximated as black whatever be the Hue or the Saturation. On the other hand, with increasing values of the Intensity, Saturation threshold that separates Hue dominance from Intensity dominance goes down.

WBC image segmentation and classification using RVM

2231

3.2. Median Filter Filtering is a part of image enhancement it is used to enhance certain details such as edges in the image that are relevant to the application. Additionally there to, filtering can even be used to eliminate unwanted components of noise. Median filtering could be a widespread technique of the image improvement for removing noise without effectively reducing the image sharpness [21].Median filter is kind of common as a result of it provides excellent noise-reduction talents, with primarily less blurring than similar size linear smoothing filters.

3.3. Image enhancement Bright stretching is a process that also used auto scaling method which is a common linear mapping function to enhance the brightness and contrast level of an image. This method is based on Equation 2 [22]. The bright stretching process is implemented based on Equation 3 [23], (2) (3) Where, TH: threshold value NewTH: bright stretching factor. In(x, y) is a value of color level at pixel (x, y) from the input image.

3.4. Image segmentation using Modified Fuzzy Possibilistic C Means The main function of the FPCM is modified that explains about the linearly weighted sum is given below equation

(4) With the constraints as follows: (5) (6) Here also the objective function is solved by iterative process where the degrees of membership, typicality and the cluster centers are update as follows

2232

S. Ravikumar and A. Shanmugam

,

(7)

,

(8)

(9) The FPCM is hybridization of PCM and FCM. The result of FPCM is better than FCM which is proved already but some of the disadvantages are solved here by this modified FPCM. The objective function of FPCM is modified by introducing histogram-based weight. This modified method achieves more desirable performance compared with standard FPCM and FCM. This proposed algorithm accounts for pixel spatial information which helps to keep continuity on neighboring pixel values of the cells.

3.5. Image classification using Relevance Vector Machine(RVM) Given a training data set of input-target pairs , RVM follows the standard probabilistic formulation and assumes that the targets are samples from the model with additive noise: (10) Where

are error terms which are generally assumed to be independent

identically distributed Gaussian variables with mean zeros and variance The likelihood function can be written as (11) Where

e is the design matrix with

, in which

, and k is a kernel Maximum likelihood estimation of w and

from (11) will generally lead

to severe over fitting, so RVM encodes a preference for smoother functions by defining an automatic relevance determination Gaussian prior over the weights: (12) The posterior over the weights is then obtained from Bayesian rule:

WBC image segmentation and classification using RVM

2233

(13) Where By integrating the weights of the product of (11) and (12): , RVM obtains the marginal likelihood for the hyper-parameters: (14) Because the values of α

that maximize the

function defined in (14) cannot be obtained in closed form, RVM considers an alternative formulae for iterative re-estimation of .

3.6. Feature selection The feature selection is done after building the feed forward network (SLFN). To improve the efficiency, three feature sets are F1, F2 & F3.  F1=(Area,Perimeter,MajorAxisLength,MinorAxisLength,Eccentricity,Orient ation,ConvexArea,EquivDiameter,Extent,MeanIntensity,MaxIntensity,Min Intensity).  F2= (Area, Perimeter, Eccentricity, Convex Area, EquivDiameter, Extent, number of nucleus lobes).  F3= (Area, Perimeter, Eccentricity, EquivDiameter, number of nucleus lobes).

4. Experimental Results To evaluate the techniques results, the experiment is conducted on various blood cell images. The blood smear image samples are collected from research laboratory in a reputed hospital in Coimbatore. The blood cell image contains RBC, WBC and platelets. From those the WBC are alone segmented and its number of WBC detected by various techniques is compared with actually present in the image which is manually obtained.

2234

S. Ravikumar and A. Shanmugam

Figure 4.1 Sample Input Blood Cell Images

Figure 4.2 RGB to HSV Output

Figure 4.4Image filtering Output

Figure 4.3Image enhancement

Figure 4.5MFPCM Segmentation output

The figure 4.1 is taken as an input image which consists of RBC, WBC and platelets in which WBC alone is clustered by various techniques. The figure 4.2 gives the result of RGB to HSV; figure 4.3 displays the image enhancement output, Figure 4.4 shows the filtering output and finally figure 4.5 gives the segmentation output of proposed MFPCM. During the testing phase, the test values are compared with the manual knowledge base for each type of blood cell. The classification efficiency is measured for the three feature sets F1, F2 and F3. Out of the 85 sample data extracted from 40 images, 62 samples are considered as training data and remaining 23 samples as testing data. Table 1 gives the comparison between proposed RVM and Existing ELM [17] for testing efficiency. In this table comparison are taken by three feature selection that are given as F1, F2 and F3. These feature selection are based on area, perimeter, convex length, diameter and number of lobes. Activation functions are taken placed for this comparison they are unipolar sigmoid, bipolar sigmoid, radial basis kernal. A testing efficiency of up to 85% is obtained for feature set F1 and around 89% in case of feature set F2. Comparing F1 and F2 sets, F3 gave the maximum efficiency up to 91% for proposed RVM. In this case of exiting ELM gives result of testing efficiency of up to 67% is obtained for feature set F1 and around 69% in case of feature set F2. Comparing F1 and F2 sets, F3 gave the maximum efficiency up to 70%. From the above table it is clearly noticed that the proposed method of RVM gives the better result than existing method of ELM [17]. Therefore overall testing efficiency of proposed RVM has taken better result in F3.

WBC image segmentation and classification using RVM

2235

Table 1. Testing efficiency Comparison Feature

F1

F2

F3

Activation Function

Testing Efficiency Proposed RVM

Existin ELM

Unipolar sigmoid

85%

67%

Bipolar sigmoid

83%

65%

Radial basis kernel

81%

63%

Unipolar sigmoid

89%

69%

Bipolar sigmoid

88%

67%

Radial basis kernel

86%

68%

Unipolar sigmoid

90%

68%

Bipolar sigmoid

91%

70%

Radial basis kernel

89%

61%

5. Conclusion The White Blood Cells is an important blood cells which becomes reasons for many diseases occurring presently. The diagnosing of WBC becomes vital process today. In this paper first the RGB image of WBC cells are changed into HSV. Filtering process are followed by HSV, here median filter are take placed because it preserves the edge of the images. Followed this filtering process image enhancement are take placed, here bright stretching methods are used. In this paper the WBC cells are taken for segmentation and efficient segmentation is achieved by using fuzzy based techniques. WBC classification is achieved by Relevance Vector Machine (RVM). These are all done and they are evaluated in MATLAB using blood Cells Images. From the comparison table it is clearly observed that the proposed method of RVM gives the best result in WBC classification than existing method. Thus this technique can be used for successful segmentation and classification of WBC.

References [1] Sordo.M([email protected]), “Introduction to Neural Networks in Healthcare,” Usenet post to http://www.openclinical.org/neuralnetworks.html, October 2004. [2] “Blood”, Keith Breden Taylor and Julian B. Schorr, Colliers Encyclopaedia, Vol 4, 1978

2236

S. Ravikumar and A. Shanmugam

[3] Vincenzo Piuri, Fabio Scotti: “Morphology Classification of Blood Leucocytes by Microscope Images”, IEEE International Conference on Computitional Intelligence for Measurement Systems and Applications, Boston, MA, USA, 14-16 July 2004. [4] Complete Blood Count, 2007 available at Wikipedia, http://en.wikipedia.org [5] Jea-Sang Park, james M. Keller (1997): “Fuzzy Patch Label Relaxationin Bone Marrow Cell Segmentation”, IEEE [6] NiponTheera-Umpon, Paul D. Gader: “System-Level Training of Neural Network for Counting White Blood Cells”; IEEE Transactionson Systems, Man and Cybernetics, Part C: Appliccations and Reviews, Vol 32, No 1, February 2002 [7] FatemehZamani, Reza Safabakhhsh: “An Unsupervised GVF Snake Approach for White Blood Cell Segmentation Based on Nucleus”, Signal Processing, The 8th International Conference on Volume 2, 2006. [8] Jea-Sang Park, james M. Keller (1997): “Fuzzy Patch Label Relaxation in Bone Marrow Cell Segmentation”, IEEE [9] Lena Costrarido, Medical Image Analysis Methods: Evaluation Strategies for Medical-image Analysis. Taylor & Francis, pp 433-471, United Stated of America, 2005. [10] FatemehZamani, Reza Safabakhhsh: “An Unsupervised GVF Snake Approach for White Blood Cell Segmentation Based on Nucleus”, Signal Processing, The 8th International Conference on Volume 2, 2006. [11] Milan Sonka, VaclawHlavac and Roger Boyle, “Image processing, Analysis and Machine Vision”, Chapman and Hall, London, 1993. [12] Ferdinand van der Heijden, “Image Based Measurement Systems, Object Recognition and Parameter Estimation”, John Wiley &Sons, West Sussex, England, 1995. [13] Khoo Boon How, Alex See Kok Bin, Alex See Kok Bin, Ng TeckSiong, Khoo Kong Soo: “Red Blood Cell Segmentation Utilizing Various Image Segmentation Techniques”, Proceeding of International Conference on Man-Machine Systems 2006, September 15-16 2006, Langkawi, Malaysia. [14] E. Montseny, P. Sobrevilla, S. Romani (2004): “A Fuzzy Approach to White Blood Cells Segmentation in Color Bone Marrow Images”, IEEE on 25-29 July, 2004, Budapest, Hungry. [15] Hengen, H , O. Gabel, A. Hajra, T. Paulus, M. Ross and S. Spoor, “Development of a system for the analysis and classification of blood and bone marrow cell images to support morphological diagnosis of leukemia,” Kaiserslautern, Germany, February 2004. http://www.eit.uni-kl.de/pandit/main/forschung/blutzellene.htm [16] Sabino, D. M. U, L. da F. Costa, E. G. Rizzatti and M. A. Zago, “A texture approach to leukocyte recognition,” Real-Time Imaging, Vol.10, pp. 205–216, Usenet post to http://www.cse.iitd.ernet.in/~csa03027, February 2004.

WBC image segmentation and classification using RVM

2237

[17] M.Alagappan, B.BanuRekha, R.Arun, M.Kalaikamal, S.Muthukrishnan, C.S.Sai Ganesh, S.Sathishkumar, “Extreme Learning Machine (Elm) Based Automated Identification And Classification of White Blood Cells”, Department of Biomedical Engineering, PSG College of Technology [18] Yang, L, P. Meer and D. J. Foran, “Unsupervised Segmentation Based on Robust Estimation and Color Active Contour Models,” IEEE, Usenet post to http://www.caip.rutgers.edu/riul/research/papers/pdf/snake.pdf, January 2004. [19] Sinha, N ([email protected]) and A. G. Ramakrishnan ([email protected]), “Automation of Differential Blood Count,” Bio-Medical Lab, Department of Electrical Engineering, Indian Institute of Science, Bangalore, IEEE, 2003. [20] Ongun,G, U. Halici, K. Leblebicioglu and V. Atalay, “Feature Extraction and Classification of Blood Cells for an Automated Differential Blood Count System,” IEEE International Joint Conference on Neural Networks, pp. 2461-2466, 2001. [21] Chan, R.H., C.W. Ho and M. Nikolova, Salt-and pepper noise removal by median-type noise detectors and detail-preserving regularization. IEEETrans. Image Proc., 14: 1479-1485. 2005 [22] R.W.Jr. Weeks,(1996). Fundamental of Electronic Image Processing. Bellingham: SPIE Press. [23] Nor HazlynaHarun ,N.R.Mokhtar ,M.Y. Mashor , H.Adilah, R.Adollah, Nazahah Mustafa, N.F.MohdNasir , H.Roseline, ‘Color image enhancement techniques based on partial contrast and contrast stretching for acute leukaemia images’, ICPE-2008.

Received: March 1, 2014

Suggest Documents