Leukemia Detection in the White Blood Cell Count using Sift Technique and Classification Jasleen Kaur Department of Computer Science SVIET, Banur Punjab, India E-mail: sippyjasleen@gmail.com Miss Amrinder Kaur Assistant Professor SVIET, Banur Punjab, India E-mail: amarinder270.sebiz@gmail.com Abstract: Leukemia Detection is planned of automatic advance. A physical technique of LEUKEMIA DETECTION, specialist checks mini images. Leukemia detection is produces in the bone marrow. Basically Leukemia is detected only by investigating the white blood cells. Attentive only on WBC, Leukemia Detection system analyses the microscopic image and overcome these problems. It removes the necessary parts of images and direct applies some techniques. K-mean collecting is used only WBC (WHITE BLOOD CELL) detection. In this thesis we describe a system for medical data processing that mainly uses Sift (scale invariant feature transformation), best solution identifies and classification has been done (Feed Forward Neural Network). The noise is removed from the image using filter. Identify the white blood cell. Feature Extraction using SIFT is the processing of converting the image into data. Input image the BPNN classifier that test image into either infected or not infection. The major part of this work is to segment the white blood cells for leukemia detection. An evaluate the performance parameters like false acceptance rate, false rejection rate, accuracy and compare the white blood cell in manual and auto count. Keywords: Leukemia Detection, WBC, BPNN classifier, SIFT, Feed Forward Neural Network I. INTRODUCTION Medical imaging has become one of the most significant conception and explanation methods in ecology and medicine over the previous decade. This time has perceived a incredible expansion of new, prevailing apparatuses for detecting, packing, conducting, analysing, and exhibiting medical images. This has led to enormous growth in the application of digital image dispensation techniques [1] for cracking medical difficulties. The microscopic images of the blood cells are experiential to find out numerous diseases. Variations in the blood condition show the development of diseases in an individual. Leukaemia can central to demise if it is left unprocessed. Based on some statistics it is found that the leukaemia is the fifth cause of death in men and sixth cause of death in women. Leukaemia originates in the bone marrow. Each bone comprises a thin substantial inside it which is also known as a bone marrow which is shown in the fig. 1. Leukemia is the cancer of the blood. It starts in the bone marrow [3], it is the area where blood cells are made. When you have leukaemia, the bone marrow starts to make a lot of abnormal white blood cubicles, called leukaemia cells. They don't do the exertion of normal white blood cells. They grow faster than normal cells, and they don't break increasing when they should. Over time, leukaemia cells can crowd out the normal blood cells. This cans chief to serious difficulties such as anaemia, bleeding, and infections. Leukaemia cells can also spread to the lymph nodes or other organs and origin bulge or pain [2]. There are numerous different types of leukaemia. In general, leukaemia is collected by how fast it gets poorer and what kind of white blood cell it affects. Figure no: 1 Bone marrow and blood component [1] The cells in the bone marrow start changing and they get infected and become leukaemia or infected cells. These leukaemia cells are having strange properties than the normal cells. RES Publication 2012 Page 166
II. TECHNIQUES & METHODS A. Segmentation using Zack Algorithm from Local Scale-Invariant Features", David G. Lowe). Each of these highlight vectors is invariant to any scaling, revolution or interpretation of the picture. According to Zack's algorithm, in grey intensity histogram (h[x]) of the enduring sub-image mechanisms, a line is created between the highest histogram value (h[bmax]) and the lowest histogram value (h[bmin]), where bmax and bmin indicate the grey level values in which the histogram h[x] influences its maximum and minimum, individually. B. k-means Clustering Algorithm Simply speaking it is a method to classify or to collection your items based on attributes/features into K number of group. K is positive digit amount. The grouping is complete by minimizing the sum of squares of distances between data and the corresponding cluster centroid. Thus, the reason of K-mean clustering is to classify the data.[17] In K-means clustering If the numeral of information is less than the numeral of cluster then we assign each data as the centroid of the cluster. Figure no 3 Scale Invariance Feature Transformation [18] This methodology offers numerous highlights with neuron reactions in primate vision. To help the extraction of these highlights the SIFT calculation applies a 4 stage separating methodology: Scale-Space Extreme Detection This phase of the separating endeavours to recognize those areas and scales that are identifiable from diverse perspectives of the same item. Key point Localisation [15] This stage endeavors to take out more focuses from the rundown of key points by discovering those that have low difference or are inadequately restricted on an edge. Figure no 2 K-mean [17] C. Sift Algorithm (Scale Invariance Feature Transformation) TECdfHNIQaxsxsxUES & MffsfrfsdETHODS Introduction Assignment This step intends to relegate a predictable introduction to the key points taking into account nearby picture properties. The SIFT approach, for picture highlight era, takes [14] a Key point Descriptor picture and changes it into an "expansive gathering of The nearby inclination information, utilized above, is neighbourhood highlight vectors" (From "Article Recognition additionally used to make key point descriptors. The slope RES Publication 2012 Page 167
data is pivoted to line up with the introduction of the key point and after that weighted by a Gaussian with change of 1.5 * key point scale. D. Feed Forward Neural Network. Feed Forward Neural Network is an organically stimulated organization algorithm. It consists of amount of simple neuron like processing units, prearranged in layers. These connections are not all equal: each joining may have a different strength or weight. The weights on these contacts encode the information of a network. Frequently the units in a neural network are also called nodes [17]. Figure no: 5 Feed Forward Neural Network The 3 inputs are shown as loops and these do not belong to any layer of the network (though the inputs occasionally are measured as a simulated layer with layer number 0). Any layer that is not an output layer is a hidden layer. This network consequently has 1 hidden layer and 1 output layer. The numeral also shows all the networks between the units in different layers. A layer only joins to the preceding layer. III. RELATED WORK W. Tang, H. Cao [1] presents Detection of different types of cancers is important in clinical diagnosis and treatment. Leukemia is one of the cancer s with different subtypes: acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML). The detection of these subtypes according to different genetic markups in leukemia patients will lead to individualized therapies. Jiang, Yong, et al. [3] The purpose of this study was to study the extract of Rumex root which had an inhibitory action on the cell proliferation of human leukemia cell line THP-1. Methods: The combination of percolation and extraction was used to separate and extract the main chemical compositions of Rumex, MIT was used to assay the curve of inhibition ratio Mohapatra, Subrajeet, et al. [2] Acute lymphoblastic leukemia (ALL) are a group of hematological neoplasia of childhood which is characterized by a large number of lymphoid blasts in the blood stream. ALL makes around 80% of childhood leukemia and it mostly occur in the age group of 3-7. The nonspecific nature of the signs and symptoms of ALL often leads to wrong diagnosis. Diagnostic confusion is also posed due to imitation of similar signs by other disorders. Careful microscopic examination of stained blood smear or bone marrow aspirate is the only way to effective diagnosis of leukemia. Victor Lavrenko [17] The K-means algorithm starts by K points(centroids)at random loactions in space,we then perform the following steps iteratively.(1) for each instance, we assignit to the cluster with the nearest centroid, and (2) we move each centroid to the mean of the instances assigned to it. The algorithm continuous until no instances change cluster membership. Singla, Savita, and Reecha Sharma [15 ] Image stitching is a technique that combines two or more images from the same scene to obtain a panoramic image. Image stitching is used in Medical system for stitching of X-ray images. As the flat panel of X - ray system cannot cover all the parts of a body. So stitching of Medical images can be done. Stitching of images basically includes two main parts Image Matching and Image Blending. For Image Matching the two algorithms SIFT and SURF are used. This paper presents a technique using hybrid of SIFT and SURF. As SIFT is slow process and not suitable at illumination changes while it is invariant to scale changes, rotation and affine transformations whereas RES Publication 2012 Page 168
SURF is having illumination property and having high computational speed. Li, Zisheng, et al [14] An effective method for quantitatively evaluating rigid and non-rigid image registration without any manual assessment is proposed. This evaluation method is based on feature point detection in reference images and corresponding point localization in registered floating images. For feature point detection, a 3D SIFT keypoint detector is applied to determine evaluation reference points in liver vessel regions of reference images. For corresponding point localization, a 3D phase-only correlation approach is applied to match reference points. IV. IMPLEMENTATION AND RESULTS There various performance parameters are used to evaluate the performance of detecting the leukemia in medical processing. The design and implementation for proposed study in MATLAB A. Design The design of the proposed work, thorough study of existing detection technique is done and compared with the proposed method. B. Performance Metrics There are different metrics are used to evaluate the performance of the detection are false acceptance rate, Accuracy, false rejection rate, Mean Square Error, Auto and Manual count etc. B.1 Mean Square Error rate The mean squared error (MSE) or mean squared deviation (MSD) of an estimator measures the average of the squares of the errors or deviations. B.2 Accuracy It is used to describe the closeness of a measurement to the true value. When the term is applied to sets of measurements of the same measured, it involves a component of random error and a component of systematic error. B.3 False Acceptance Rate The False Acceptance rate (FAR) is the probability that the system incorrectly authorizes a non-authorized person, due to incorrectly matching the biometric input with a template. B.4 False Rejection Rate The FRR or False Rejection Rate is the probability that the system incorrectly rejects access to an authorized person, due to failing to match the biometric input with a template. Image Table no. 1 Previous RES Publication 2012 Page 169 no Manual Count Exiting work Auto count Existing Img 1 12 11 91 Accuracy in Existing Img 2 18 18 100 Img 3 15 12 80 Img 4 8 8 100 Img 5 24 19 79 Img 6 5 5 100 Img 7 2 2 100 Image no Table no: 2 Manual Count work Auto count Accuracy Img 1 19 1 98.7 Img 2 23 3 97.4 Img 3 15 6 99.1 Img 4 4 8 98.0 Img 5 15 2 99.9 Img 6 18 4 97.8 Img 7 22 1 99.2 in
V. CONCLUSION AND FUTURE SCOPE Figure no: 5.7.4 Comparison between Manual count in Main center of this paper is to research an automated system which can detect the leukemia from the microscopic image to improve the accuracy and reduce the time to detect than the manual approach.so many lives can be save by using the proposed approach of leukemia detection. The main part of this work is to segment the white blood cell for leukaemia detect. The accuracy gets proposed system is 99.89%.We can also use the purposed system to find out the percentage of leukaemia infection in microscope image. We can improve the segmentation scheme which can segment the overlapped cells also. There were found the use of optimization techniques in some systems. We can also use Firefly optimization technique to improve the auto count of the reduction. Doing so will increase the cost but accuracy will also be improved. REFERENCE Figure no: 5.7.5 Comparison between Auto count in Figure no: 5.7.6 Comparison between Accuracy in The above figure shows the result is shows in the table 1. The accuracy found the existing system is 93.57%. The existing system also counts the percentage of the infection in the blood image.the microscopic images result is shown in the table 2. The accuracy found the proposed system is 98.77%. The proposed system also counts the percentage of the infection in the blood image. [1]. [1] W. Tang, H. Cao, and Y.-P. Wang, Subtyping of Leukemia with Gene Expression Analysis Using Compressive Sensing Method, 2011 IEEE First Int. Conf. Healthc. Informatics, Imaging Syst. Biol., no. 1, pp. 76 80, 2011. [2]. [2] Mohapatra, Subrajeet, et al. "Fuzzy based blood image segmentation for automated leukemia detection." Devices and Communications (ICDeCom), 2011 International Conference on. IEEE, 2011 [3]. [3] Jiang, Yong, et al. "Extractives of Rumex restrain the proliferation of leukemia cell THP-1." IT in Medicine and Education (ITME), 2011 International Symposium on. Vol. 2. IEEE, 2011 [4]. [14] Li, Zisheng, et al. "Evaluation of medical image registration by using 3D SIFT and phase-only correlation." Abdominal Imaging. Computational and Clinical Applications. Springer Berlin Heidelberg, 2012. 255-264. [5]. [15] Singla, Savita, and Reecha Sharma. "Medical Image Stitching Using Hybrid Of Sift & Surf Techniques." International Journal of Advanced Research in Electronics and Communication Engineering (IJARECE) Volume 3, Issue 8, August 2014 [6]. [20] Schuldt, Christian, Ivan Laptev, and Barbara Caputo. "Recognizing human actions: a local SVM approach." In Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on, vol. 3, pp. 32-36. IEEE, 2004. [7]. [21] Madhloom HT, Kareem SA, Ariffin H, Zaidan AA, Alanazi HO, Zaidan BB. Anautomated white blood cell nucleus localization and segmentation using imagearithmetic and automated threshold. J Appl Sci 2010;10(11):959 66. RES Publication 2012 Page 170