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.
Related Documents
International Journal of Innovative Research in Advanced Engineering (IJIRAE) ISSN: 2349-2163 Volume 1 Issue 8 (September 2014) _________________________________________________________________________________________________ © 2014, IJIRAE- All Rights Reserved
    International Journal of Innovative Research in Advanced Engineering (IJIRAE) ISSN: 2349-2163   Volume 1 Issue 8 (September 2014 ) _________________________________________________________________________________________________ © 2014, IJIRAE- All Rights Reserved Page -301   TEXTURE ANALYSIS TO DETECT MALARIA TROPICA IN BLOOD SMEARS IMAGE USING SUPPORT VECTOR MACHINE    Abstract— Malaria tropical disease caused by protozoan parasites of the genus Plasmodium that infect the red blood cells of  patients. Process is performed by physicians in diagnosing diseases of tropical malaria of blood smear with directly observed  through a microscope. This procedure is certainly not effective. Therefore, the detection of tropical malaria automatically by  observing the shape and pattern of the image of blood smears will be very helpful. The aim of our study to detect a disease of  tropical malaria in blood smears image based on texture analysis. This study consists of two steps. The first step is segmentation of  blood smear image using active contour method. The second step is detection of tropical malaria by performing classification of  suspected blood smears containing plasmodium malaria using support vector machine (SVM). The results of the testing that has  been done for normal blood obtained accuracy is 93%, phase thropozoit with accuracy 92.5%, Schizont phase with accuracy 98.3%,  and phase gamethozit with accuracy 100%.  Keywords— Active contour, support vector machine, tropical malaria. I.   I NTRODUCTION   Malaria is a disease caused by protozoan parasites of genus plasmodium that infect red blood cells of patients. Parasites enter human body through bite of a female Anopheles mosquito. Plasmodium species that infect humans, namely plasmodium falcifarum, plasmodium vivax, plasmodium ovale, and plasmodium malariae. Plasmodium vivax causes tertian malaria, plasmodium malaria is cause of malaria kuartana, plasmodium ovale causes ovale malaria, while  plasmodium falciparum causes tropical malaria [1]. Falciparum malaria in tropical or tropical malaria is the most severe form of malaria. Characterized by irregular heat, anemia, splenomegaly, and frequent parasitaemia complications. Incubation period is 9-14 days. Tropical malaria attack all forms of erythrocytes. Caused by Plasmodium falciparum. Shape of this Plasmodium is ring or small ring. Has a diameter 1/3 of normal diameter of erythrocytes, and only species that has 2 core chromatin [2]. Diagnosis can be done if malaria parasite found in blood of patients. Method is performed by physicians in diagnosing diseases of tropical malaria on blood smears is to observe directly through use microscope. From observation with microscope will be known who is image of blood smears of healthy red blood cells, red blood cells in which there is plasmodium falciparum with different phases, such as thropozoit phase, schizont phase, and gametozide phase located outside the red blood cells [2]. Detection manually will consume lot of time, other than that doctors may differ in defining image of a normal blood  preparations and infected. Thus detection of tropical malaria automatically by observing shape and pattern of image of  blood smears will be very helpful. Previous studies related to detection of tropical malaria automatically through blood smears image has been done [3]–[7]. Purpose of study was to detect automatically of tropical malaria on blood smears image using a support vector machine (SVM). There are two main steps in this research. First step is segmentation blood smears image using active contour method [8] which resulted in area being treated as a red blood cell, and second step is detection of tropical malaria by performing classification on image of suspected disease, using support vector machine (SVM) [9]. Support vector machine method used in this study are based on binary svm one against all [10]. Features used are statistics, Gray Level Co-occurrence Matrix (GLCM) and Gray Level Run Length Matrix (GLRLM) [11]. The results of testing that has been done for normal blood obtained accuracy is 93%, thropozoit phase with accuracy 92.5%, Schizont phase with accuracy 98.3%, and gamethozit phase with accuracy 100%. II.   MATERIAL   AND   METHODS  A.    Material Input image used in this study is image of blood smears taken from Consists of two data types, namely image used for training data and test data. Image of blood used for training data consists of 120 images, to images representing each class number 30, each of which has a size of 50x50 pixels.  B.    Blood Smear Segmentation Purpose of segmentation is to divide image into several parts, in order to present image becomes more meaningful and easier to perform further analysis.   Sri Widodo    Medical records and Health Information  Departement,APIKES Citra Medika Surakarta Central Java Indonesia Wijiyanto  Department of Informatic Systems, STMIK  Duta Bangsa Surakarta, Central Java,  Indonesia    International Journal of Innovative Research in Advanced Engineering (IJIRAE) ISSN: 2349-2163   Volume 1 Issue 8 (September 2014 ) _________________________________________________________________________________________________ © 2014, IJIRAE- All Rights Reserved Page -302   In process of image segmentation blood preparations with 256x256 pixel size on image to be tested, will produce images of blood preparations which will be used later in the process with the size of 50x50 pixels. Process of blood smears segmentation using active contour without edges method [12], [13]. Steps in segmenting: 1.   Initial process (preprocessing) is to change image into a grayscale image (256x256 pixels). 2.   Segmentation using active contour without edge: a.   Initializing beginning of active contour C={(x,y):φ(x,y)=0} ∀ (x,y) ∈   Ω   b.   Calculating value of u and v, which represent interior and exterior regions c.   Calculating energy of contour function FF=(I-v) 2 -  λ 2 (I-u) 2  d.   Calculating of flow e.   An update to contour f.   Perform steps b, c, d and e until convergence condition. 3.   Save segmentation results into a file, with size of 50x50 pixels. Fig. 1 Steps of proposed method for detecting malaria tropical C.   Feature    Extraction   In feature extraction stage of image data of blood preparations is process to get accurate information so that identification process can be done. Features used are statistical features which include standard deviation, mean, skewness, entropy, kurtosis and grayscale images, features Gray Level Co-occurrence Matrix (GLCM) is mean, entropy, standard deviation, variance, correlation, energy, homogeneity,and Gray Level Run Length Matrix (GLRLM) which includes: Short Runs Emphasis (SRE), Long Runs Emphasis (LRE), Gray Level Non-uniformity (GLN), Run Percentage (RP), Run Length Non-uniformity (RLN), low Gray Level Run Emphasis (LGRE), High Gray Level Run Emphasis (HGRE) [11].  D.   Tropical Malaria Detection Using SVM The next process is detection of malaria by doing classification using Support Vector Machine (SVM) [14]. Concept of SVM can be explained simply as an attempt to find the best hyperplane that serves as a separator of two classes in input space. Hyperplane in a d-dimensional vector space is an affine subspace dimension d-1 which divides vector space into two parts, each of which corresponds to a different class. Figure 2 shows some of patterns that are members of two classes: +1 and -1. Pattern belonging to class -1 symbolized by red color (box), while pattern in class +1, symbolized by yellow color (circle). Classification problem can be translated to business of finding a line (hyperplane) that separates two groups. Various alternative dividing line (discrimination boundaries) are shown in Figure 2 (a).    International Journal of Innovative Research in Advanced Engineering (IJIRAE) ISSN: 2349-2163   Volume 1 Issue 8 (September 2014 ) _________________________________________________________________________________________________ © 2014, IJIRAE- All Rights Reserved Page -303   The best separation hyperplane between two classes can be found by measuring hyperplane margin, and seek its maximum extent. Margin is distance between hyperplane to closest pattern of each class [15]. In this study using multiclass SVM one against all. This method will build a binary SVM k, where k is the number of classes [10]. Ith SVM was trained with all the samples in ith class with positive class label and all other samples with negative class labels. Examples of methods for classification of one against all four classes, illustrated in figure 3 and table 1. In image shown there are 3 decision function, third function of decision-making function is maximum taken. Fig. 2 Hyperplane that separates two classes (-1 and +1) Table 1 Example of one against all method     y i  = 1  y i  = - 1 Hypothesis Class 1 Not Class 1  f 1 (x) = (w 1 )x + b 1 Class 1 Not Class 1  f 2 (x) = (w 2 )x + b 2  Class 1 Not Class 1  f 3 (x) = (w 3 )x + b 3  Class 1 Not Class 1  f 4 (x) = (w 4 )x + b 4   Fig. 3 Classification method of svm ones against ones In classification process of training data, variable hyperplane for each classifier (classifier) gained will be saved and will be used as data for each classifier in testing process, in other words, the training data classification process is to seek support vector, alpha and bias of input data (in this case used quadratic programming). While in testing process, image of  blood preparations used is blood smears which had never been trained in training process with a size of 256 x 256 pixels. Segmentation process is done using active contour cropping to get image of size of 50 x 50 pixels which will be used for classification process. For testing process using results of data feature extraction and classification results of data training.   Results of this process is index value of the largest decision function, stating class of testing data. If a class in classification test match test data classes, classification is stated correctly. Final result of classification is image of blood that matches with index value of decision function using SVM one against all . Basis of decision-making using a multiclass SVM is SVM Binner. Based on classification process using SVM, classification is divided into two parts: training and testing. Here is a training algorithm for each binary SVM: input is Xtrain matrix (matrix of feature extraction results of training) and Ytrain vectors as a pair of target input and its output is w, x, b (variables hyperplane equations). Steps are described as follows:    International Journal of Innovative Research in Advanced Engineering (IJIRAE) ISSN: 2349-2163   Volume 1 Issue 8 (September 2014 ) _________________________________________________________________________________________________ © 2014, IJIRAE- All Rights Reserved Page -304      Determine input (Z = Xtrain) and Target (Ytrain) as a pair of training from two classes.    Calculating Gaussian kernel K(Z,Z  i ) = exp    Calculate Hessian matrix  H = K(Z,Z  i ) * Y * Y  T       Assign c and epsilon.    Assign vector e as unit vector which has same dimension with dimension of Y.    Calculating quadratic programming solution:min  ,  with   dan   Input for matrix Z is a matrix of features resulting from feature extraction process and Ytrain vector as a target. Ytrain vector is column vector for classification of first class, where all images of blood preparations of first class will be symbolized by number 1, all images of blood smears from other classes with number -1. In this study, used a Gaussian kernel function with variance (σ) = 1.  The next step is to calculate Hessian matrix, ie multiplication of a Gaussian kernel with Ytrain. Ytrain here is a vector that contains values 1 and -1. Hessian matrix is later used as input variables in quadratic programming. In this study using quadratic programming monqp function of SVM-KM tool. Monqp  programming quadratic function requires variables of c and epsilon. Therefore, determine value of c and epsilon, c is a constant in Lagrangian multipliers and epsilon (cost parameter) is upper limit val ue of α which serves to control  classification error. Unit vector e is also formed with dimensions equal to vector of Y. Completion   min with quadratic programming, an implementation of the search for solutions to the problems . This study used value of c  = 100000 and epsilon  = 1x10 -7 . Results from monqp function (quadratic programming) is variable value of w, x, and b, that will be used for testing process. III.   C ALCULATING OF   ACCURACY,   SENSITIVITY   AND   SPESIFICITY   To calculate accuracy, sensitivity and specificity using Receiver Operating Characteristics (ROC). First, results of classification will be done so that comparison will be obtained four values, each of which is a true positive, false negative, false positive, and true negative. True positive (TP) shows image of blood mears correctly identified according to the class (V). False positive (FP) is image of blood smears that should be identified with class turned out to be right in  process of identifying wrong classification. True negative (TN) is an image that is not a member of class identified right is not a member of class (NV). False negative (FN) shows image of blood smears that should not members of class identified as a member of class. IV.   EXPERIMENT   AND   RESULTS Tests on SVM classification method using 160 dataof blood smears images with size 50x50 pixels. Each class of classification consists of 40 data. In a trial carried out a two-steps process. The first step is training phase, while second is test phase. The training phase is used to obtain coordinates of support vector, weight, bias and suppot distance vector, whereas testing phase is to use data other than training data to obtain the results of classification, so as to know level of accuracy. The testing process is done in four classes (normal blood, blood thropozoit, blood schizont and gametozit). Each group consists of 40 data. (a) (b) Figure 4. Segmentation process of blood smears (a)   Process of segmentation (b) Normalized form  
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