Artigo

Publish in

Documents

22 views

Please download to get full document.

View again

of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Share
Description
2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012 Segmentation Based Approach for Detection of Malaria Parasites Using Moving K-Means Clustering A.S.Abdul Nasir, M.Y.Mashor Electronic & Biomedical Intelligent Systems (EBItS) Research Group, School of Mechatronic Engineering, Universiti Malaysia Perlis, Campus Pauh Putra, 02600 Pauh, Perlis, Malaysia. Email: aimi_salihah@yahoo.com Abstract—Recent progress based on microscopic im
Tags
Transcript
  Segmentation Based Approach for Detection of Malaria Parasites Using Moving K-Means Clustering A.S.Abdul Nasir, M.Y.Mashor Electronic & Biomedical Intelligent Systems (EBItS) Research Group, School of Mechatronic Engineering, Universiti Malaysia Perlis, Campus Pauh Putra, 02600 Pauh, Perlis, Malaysia. Email: aimi_salihah@yahoo.com Z.Mohamed Department of Medical Microbiology & Parasitology, School of Medical Sciences, Health Campus, Universiti Sains Malaysia, 16150 Kubang Kerian, Kelantan, Malaysia. Email: zeehaida@kck.usm.my   Abstract   —Recent progress based on microscopic imaging has given significant contribution in diagnosis of malaria infection based on blood images. Due to the requirement of prompt and accurate diagnosis of malaria, the current study has proposed an unsupervised colour image segmentation of malaria parasites using moving k-means (MKM) clustering algorithm. It has been applied on malaria images of  P. vivax   species. The proposed segmentation method provides a basic step for detection of the presence of malaria parasites in thin blood smears. With the aim of obtaining the fully segmented red blood cells infected with malaria parasites, the malaria images will firstly enhanced by using the partial contrast stretching technique. Then, the MKM clustering algorithm has been applied on the saturation and intensity components of HSI (hue, saturation, intensity) colour space for segmenting the infected cell from the background. After that, the segmented images have been processed using median filter and seeded region growing area extraction algorithms for smoothing the image and removing any unwanted regions from the image, respectively. Finally, the holes inside the infected cell are filled by applying region filling based on morphological reconstruction algorithm. The proposed segmentation method has been analyzed using 100 malaria images which consist of the trophozoite and gametocyte stages. Overall, the results indicate that MKM clustering that has been performed on saturation component image has produced the best segmentation performance with segmentation accuracy of 99.49% compared to the intensity component image with segmentation accuracy of 98.89%.  Keywords-Malaria; colour segmentation; partial contrast stretching; HSI colour space; moving k-means clustering I.   I  NTRODUCTION  Malaria is a serious infectious diseases, causing wide spread sufferings and deaths particularly in Africa and south Asia. This disease has caused the death of an estimated 655,000 people in 2010, with 86% of the victims are children under five years of age [1]. Up to this date, there are five species of genus  Plasmodium  that can cause malaria have been discovered namely  P. falciparum ,  P. vivax ,  P. ovale ,  P. malariae  and  P. knowlesi  [2]. Here, the life-cycle for each species can be divided into four different stages which are ring, trophozoite, schizont and gametocyte [3]. The accurate and timely diagnoses of malaria infection are the main keys to control and cure this disease effectively. Although there are a number of alternative methods which associated with malaria diagnosis have been developed in recent years, manual microscopy examination of blood slides especially based on the thin blood smears is still widely used in most medical laboratory around the world [3]. Detection of the presence of malaria parasites in the examined blood slide is a very important task in malaria diagnosis [4]. During this process, the presence of the  parasites can be recognized based on their physical features. In addition, observation of the form of the red blood cells (RBCs) that have been infected by the parasites is also required for this  process [3]. Here, visual detection of the parasites is possible  by using the Giemsa stain [3]. The staining process highlights the parasites but slightly colourized the RBCs. Even though this process highlights the appearance of the parasites, other components such as white blood cells (WBCs), platelets as well as the artefacts have also been highlighted. Thus, the  process for malaria detection requires an ability for differentiating between the malaria parasites or infected RBCs with these non-parasitic stained components (normal RBCs, WBCs, platelets and artefacts) using visual information. In order to detect the malaria parasites, one of the main tasks that need to be performed during image processing is the segmentation of malaria image. Dealing with this important task, various conventional image processing techniques such as thresholding [5][6], edge detection [7] and morphological approach [8] have been used for segmentation of malaria  parasites. Ross et al.  [6] have proposed a thresholding based on image histogram method to identify the RBCs and possible  parasites present on microscopic slide. This method requires the determination of two threshold levels from the histogram. The first threshold is selected to separate the RBCs from the  background by using a method that maximizes the separability of the resultant classes of the gray level histogram. Meanwhile, the second threshold is selected to find the  parasites present in the image. The infected cells are then identified by morphologically reconstructing the RBC mask with the valid parasite marker. Somasekar et al.  [9] have proposed an automatic  procedure for detection of malaria parasites which include 978-1-4673-1666-8/12/$31.00 ©2012 IEEE2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012 653  255255 minTHmaxTH    minTHNmaxTH  Compression process Stretching process Compression process image pre-processing, extraction of infected cells and morphological operation steps. The extraction of infected cells has been applied using the grayscale image. The final evaluation of parasites detection using 76 images which consist of both malaria and normal blood images have  produced the sensitivity of 94.87% and specificity of 97.30%. In some cases, the proposed method indicated that the result is  positive even though the RBC is not infected with parasite. These false diagnosis results could be caused due to the WBCs or artefacts which carry the same colour as the parasites. Several segmentation techniques for malaria parasites that have been proposed might provide good result if there are differences in terms of colour between the parasites or infected cells with the background region. In addition, various research studies have developed segmentation techniques that have  been performed based on the gray level image. The current study will utilize the potential of colour image segmentation approach using HSI colour space and moving k-means (MKM) clustering algorithm for obtaining the fully segmented RBCs infected with malaria parasites based on the thin blood smear images. II.   M ETHODOLOGY  The proposed procedures for segmentation of malaria  parasites are summarized as follows: Step 1 : Step 2 : Step 3 : Step 4 : Step 5 : Step 6 : Step 7 : Capture the malaria slide images. Apply the contrast enhancement technique namely  partial contrast stretching technique on srcinal malaria image. Transform the RGB (red, green, blue) colour space into HSI colour space. Apply the unsupervised segmentation technique namely moving k-means clustering. Apply the 7×7 pixels median filter for smoothing the region of the segmented image. Apply the seeded region growing area extraction (SRGAE) algorithm [10] for obtaining the area in terms of pixels for the segmented infected cell. Fill the holes inside the segmented infected cell by applying region filling based on morphological reconstruction algorithm [11]. Further details for the image acquisition, contrast enhancement as well as the segmentation of malaria parasites using HSI colour space and MKM clustering algorithm are discussed in the following sections.  A.    Image Acquisition The first step is to acquire the images of malaria samples. In this study, the malaria images of trophozoite and gametocyte stages have been captured from the thin blood smears of  P. vivax  samples. The malaria slides are prepared by Medical Microbiology and Parasitology Department, Hospital University Science Malaysia (HUSM). The malaria slides are analyzed using 100X oil immersion objective of Leica DLMA microscope. The images are then captured in the BMP format at an image setting of 800 x 600 pixels using the Infinity-2 digital camera.  B.   Contrast Enhancement Using Partial Contrast Stretching Technique Partial contrast stretching (PCS) is a well known technique utilized to increase the contrast for overall pixels in an image. Detail descriptions of PCS technique can be referred in [12][13]. Fig. 1 briefly illustrates the two important processes during applying the PCS technique on malaria image namely the stretching and compression processes. By applying this technique, the pixels within the range of lower threshold value, minTH   and upper threshold value, maxTH   will be mapped to a new range and stretched linearly to a wider range of pixels within the new lower stretching value,  NminTH   and new upper stretching value,  NmaxTH  . On the other hand, the remaining pixels will perform the compression process. Figure 1. Partial contrast stretching process. C.    Malaria Parasites Detection Based on HSI Colour Space  In this study, segmentation of malaria parasites is  performed based on HSI colour space so that the colour content of an image can be utilized. In addition, the HSI colour space is more align in the way human interpret colour and provides better colour representation compared to RGB colour space. The conversion from RGB to HSI colour space can be computed using the following equation [14]: ⎩⎨⎧ −°= θ θ  360  Hue if if  G BG B >≤  (1) ( ) ( ) [ ] ( ) ( )( ) [ ]  ⎪⎭⎪⎬⎫⎪⎩⎪⎨⎧ −−+− −+−=  − 2121 21cos  BG B RG R  B RG R θ   (2) ( )  BG R BG RSaturation ,,min31 ++−=  (3) 00 2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012 654    ( )  BG R Intensity  ++= 31  (4) Hue is a colour attribute that describes a pure colour, while saturation gives a measure of degree to which a pure colour is diluted by white colour [14]. The intensity expresses the  brightness of the hue and saturation. Based on the above conversion, since the hue, saturation and intensity are independent of one another, each colour component can be  processed separately without worrying about the correlation among them [15]. In addition, the HSI colour space has the advantage of both colour and grayscale information in an image, thus making it suitable for many applications of gray level processing techniques. Fig. 2 shows a sample of malaria image and its colour components based on HSI colour space. Based on the qualitative analysis of several malaria images, the appearance of the infected cell is mostly highlighted in S and I components images compared to the H component image. Due to the similar appearance between the infected cell and normal RBCs regions as shown in Fig. 2(b), it would be a difficult task to segment the infected cell from the normal RBCs. Thus, S and I components have been chosen to be furthered segmented using MKM clustering. (a) Malaria image (b) Hue (c) Saturation (d) Intensity Figure 2. The three colour components of HSI colour space that have been extracted from malaria image.  D.    Image Segmentation Using Moving K-Means Clustering   A typical malaria image consists of three main regions namely parasites or infected RBCs, normal RBCs and  background regions as can be seen in Fig. 2(a). However, segmenting the parasites from the background is not an easy task for malaria image. This is due to the colour of the  parasites and the RBCs which heavily depends on the buffer solution used during the preparation of the malaria slide [16]. In addition, conventional segmentation technique for example the manual thresholding might be failed to produce a good segmented image especially in the case when the histogram of malaria image does not have distinct valleys. Therefore, an adaptive unsupervised segmentation technique is required for easily segmenting the infected cell from the background. Based on this argument, the current study will utilize the  potential of moving k-means clustering algorithm that had  been introduced by Mashor [17]. In the current study, the MKM clustering is used for segmentation of malaria parasites because it offers several advantages over the standard k-means clustering algorithm such as its capability to minimize dead centres and centre redundancy problems [17]. Furthermore, the MKM clustering has been proven to be effective in avoiding the centre from  being trapped in local minima. Here, the concept of fitness has  been introduced to ensure that each cluster has a significant number of members and final fitness values before the new  position of the cluster is calculated. Consider a malaria image of  X   × Y   pixels that will be clustered into n c  regions. Let  p (  x ,  y ) as an input pixel to be clustered and c  j  is the  j -th centre (cluster) (  x  = 1, 2, 3, …,  X  ,  y = 1, 2, 3, …, Y   and  j  = 1, 2, 3, …, n c ). For segmentation of malaria image, the number of cluster,  j  has been set to 3. The MKM clustering algorithm [17] for image segmentation can  be implemented as follows: 1.   Initialize the centres and α 0 , and set α a  = α b  = α 0  (where α 0  is a small constant value, 0 < α 0  < 1/3 and should be chosen to be inversely proportional to the number of centres). 2.   Assign all pixels to the nearest centre and calculate the centre positions using: ∑∑ ∈ ∈ =  jj c yc x j j  y x pnc ),(1 (5) 3.   Check the fitness of each centre using: ( )  ( ) 2 ),( ∑∑ ∈ ∈ −=  jj c yc x j j c y x pc f   (6) 4.   Find c  s  and c l  , the centre that has the smallest and the largest value of  f  ( . ). 5.   If  f  ( c  s ) < α a  f  ( c l  ),   a.   Assign the members of c l   to c  s  if  p (  x ,  y ) < c l  , where  x,y ∈ c l  , and leave the rest of the members to c l  .  b.   Recalculate the positions of c  s   and c l    according to: ∑∑ ∈ ∈ =  ss c yc x s s  y x pnc ),(1 (7) ∑∑ ∈ ∈ = ll  c yc xl l   y x pnc ),(1 (8) 2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012 655   Note: c  s  will give up its members before step 5a and, n s  and n l   in (7) and (8) are the number of the new members of c  s  and c l   respectively, after the reassigning process in step 5a. 6.   Update α a  according to α a  = α a  - α a  / n c  and repeat steps 4 and 5 until  f  ( c  s ) ≥   α a  f  ( c l  ). 7.   Reassign all pixels to the nearest centre and recalculate the centre positions using (5). 8.   Update α a  and α b  according to α a  = α 0  and α b  = α b  – α b  / n c  respectively, and repeat steps 3 to 7 until  f  ( c  s ) ≥   α b  f  ( c l  ). III.   R  ESULTS A  ND D ISCUSSIONS  In this study, comparisons between the S and I components of HSI colour space that have been used as input images for MKM clustering have been conducted for recognizing the significance of applying each colour component for segmentation of malaria image. In order to validate the proposed segmentation method, the analyses have  been conducted using 100 malaria images. The quality of segmented image has been determined based on both qualitative and quantitative evaluations. Fig. 3(a) shows the srcinal malaria image of gametocyte stage named as Gametocyte_1 image. Based on this malaria image, the morphologies of gametocyte are hardly be seen due to the  blurred and low image contrast. The results obtained after applying the proposed segmentation method on Gametocyte_1 image are shown in Fig. 3(b)-(j). (a) Original image (b) Partial contrast stretching image (c) S component on partial contrast stretching image (d) I component on partial contrast stretching image (e) MKM clustering on S component image (f) MKM clustering on I component image (g) Median filter for S component (h) Median filter for I component (i) SRGAE and region filling for S component (j) SRGAE and region filling for I component Figure 3. Original Gametocyte_1 image and result of images after applying the proposed segmentation method. The resultant image after applying the partial contrast stretching technique is shown in Fig. 3(b). Based on this resultant image, the contrast of the infected cell, RBCs and  background regions has been improved compared to the contrast of the srcinal image. In order to perform segmentation on malaria image, both S and I components have  been extracted from partial contrast stretching image as shown in Fig. 3(c) and (d), respectively. By referring to the S component image, it can be seen that the infected cell region appears brighter in the image, while the RBCs and background regions appear darker in the image. As for the I component image, the infected cell region appears darker in the image, while the RBCs and background regions appear as the medium and brighter parts in the image, respectively. Then, MKM clustering has been applied on both colour components for easily segmenting the infected cell from the RBCs and  background regions. Fig. 3(e) and (f) show the resultant images after applying MKM clustering on S and I components images, respectively. A good clustering algorithm should not only be able to cluster the malaria image into three different regions, but it also needs to preserve the dimensional features such as the size and shape of the infected cell as these features are important for malaria  parasites detection. Noted from Fig. 3(e) and (f), MKM clustering is able to segment the malaria image with less holes in the infected cell area, as well as less artefacts in the segmented image by using the S component compared to the resultant image provided by I component. The output image of MKM clustering has been processed using median filter for producing a cleaned and smoother image as shown in Fig. 3(g) and (h). However, there are some unwanted regions such as artefacts are still appeared on the image due to its size in which cannot be cleaned by using the 7×7 pixels median filter. Therefore, the malaria image has  been furthered processed using SRGAE algorithm for obtaining the size of the segmented region. During applying the SRGAE algorithm, any segmented regions that less than 5000 pixels are categorized as non-parasite. Thus, these non- 2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences | Langkawi | 17th - 19th December 2012 656
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks