of 9
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
  Biomedical & Pharmacology Journal  Vol. 9 (2),   663-671 (2016) Cervical Cancer Detection andClassification Using Texture Analysis M.K. SOUMYA, K. SNEHA   and C. ARUNVINODH Department of Computer Science & Engineering,Royal College of Engineering & Technology, Akkikavu , Calicut University, India*Corresponding author E-mail: soumiya.aravind@gmail.com April 13, 2016; accepted: June 16, 2016) ABSTRACT Cervical cancer is one of the deadliest cancer among women. The main problem withcervical cancer is that it cannot be identified in its early stages since it doesn’t show any symptomsuntil the final stages. Therefore the accurate staging will help to give the accurate treatment volumeto the patient. Some diagnosing tools like X-ray, CT, MRI, etc. can be used with image processingtechniques to get the staging of disease. Transform features such as contourlet and Gaborfeatures mainly based on energy are used for the prediction of output. Second-order statisticalfeatures based on contrast, correlation, energy and homogeneity are significantly used to predictoutcome from pre-treatment MR images of cervical cancer tumors. This paper proposes aclassification technique using Magnetic Resonance Images(MRI) to obtain the staging of cervicalcancer patients. Key words: Cervical cancer, Magnetic resonance imaging (MRI),Texture analysis, GLCM, Gabor, Contourlet, classification, SVM. INTRODUCTION Cervical cancer has become one of themajor causes of cancer death among womenworldwide. This can be cured in its earlier stage.For most of the cases it shows symptoms only inthe advanced stages. Cervical cancer is a cancerarising from the cervix. Cancer is due to theabnormal growth of cells that have the ability toinvade or spread to other parts of the body. Studieshave found that infection with the virus called HPV(Human papillomavirus) 1  is the cause of almost allcervical cancers.Magnetic Resonance Imaging (MRI) is awidely-used method of high quality medicalimaging. The soft tissue contrast and non-invasiveness are the important advantages of MRIs.The radiologist examined the MRI to identify thepresence of tumor abnormal tissues. The shortageof radiologist, the time conception and themisinterpretations caused by viewing the MRIthrough the naked eye called out for an automatedsystem to analyze and classify medical images. Thispaper proposes a classification model with MRImages, with the help of texture analysis techniquesand support vector machine (SVM) classification.In the paper titled New Features ofCervical Cells for Cervical Cancer DiagnosticSystem Using Neural Network. by Mustafa et al  . 3 , ithas been stated that though Pap test is the mostpopular and effective test for cervical cancer, Paptest does not always produce good diagnosticperformance. In the paper Preprocessing forAutomating Early Detection of Cervical Cancer, byDebasis Bhattacharyya et al  . 4  states that InCervigram, cervix region occupies about half of theraw cervigram image. Other parts of the imagecontain inconsequential information. This irrelevant  664SOUMYA et al. , Biomed. & Pharmacol. J., Vol. 9 (2), 663-671 (2016)information can muddle automatic identification ofthe tissues within the cervix. Asselin et al., 5  discussthe imaging methods available to provideappropriate biomarkers of tumor structure andfunction using selective regions of interest (ROI),Cluster analysis and Histogram analysis. TuridTorheim et al., 6  present the paper with textureanalysis methods and classification by using SVMto identify the cured and relapsed images. S.Jagadeeswari and S. Malarkhodi 7   presented apaper on classification by using an Artificial NeuralNetwork to identify the normal and abnormal tumorimages with Fourier transform and Gaussian lowpass filter. Rupinderpal and Rajneet 8  presented anoise removal method using discrete wavelettransform. Texture analysis Texture analysis 19  describes a variety ofimage-analysis techniques that quantify thevariation in surface intensity or patterns. Some areimperceptible to the human visual system. Textureanalysis may be particularly well-suited for thelongitudinal monitoring of disease or recovery andfor lesion segmentation and characterization.Texture analysis Approaches are usually classifiedintoãstructural,ãstatistical,ãmodel-based andãtransform methods.The structural approaches representtexture by well-defined primitives and a hierarchyof spatial arrangements of those primitives. Fordescribing the texture, one must define theprimitives and the placement rules. Providing anhonest symbolic description of the image is the mainadvantage of the this approach This feature is moreuseful for synthesis than analysis tasks.Statistical approaches do not arrange toperceive expressly the hierarchical structure of thetexture They represent the texture indirectly withthe non-deterministic properties that govern thedistributions and relationships between the greylevels of an image. Most popular second-orderstatistical features for texture analysis are derivedfrom the co-occurrence matrix. They weredemonstrated to feature a potential for effectivetexture discrimination in biomedical. The approachsupported multidimensional co-occurrencematrices was recently shown to surpass waveletpackets (a transform-based technique) whenapplied to texture classification.The model based texture analysis usingfractal and stochastic models, attempt to interpretan image texture by use of, generative image modeland stochastic model. Parameters of the model areestimated and then used for image analysis. Thefractal model 17  has been shown to be useful formodeling some natural textures. It can be used alsofor texture analysis and discrimination.Transform methods of texture analysisrepresent an image in a space whose co-ordinatesystem has an interpretation which is closely relatedto the characteristics of a texture (such as frequencyor size). Examples are Fourier ,Gabor and wavelettransforms. Due to the lack of spatial localization,methods based on the Fourier transform performpoor in practice. Gabor filters provide means forbetter spatial localization. Proposed method Magnetic Resonance Images (MRI)[9] oflocally advanced cervical cancer patients are takenfor this work. The block diagram of the proposedmethod is shown in fig 1. These are preprocessedusing contrast enhancement and segmentationtechniques along with the morphological operationssuch as erosion and dilation. Features are thenextracted by first order statistical features andsecond order grey level co-occurrence matrix(GLCM). Finally the SVM classifier will predict thetreatment outcome of the patient is cured orrelapsed. MRI Dataset Cervical cancer is the cancer arising fromthe cervix which is the lower part of the uterus. Thediagnosing tools for this purpose include Pap test,Biopsy, Computed tomography (CT or CAT) scan,Magnetic resonance imaging (MRI), Positronemission tomography (PET) scan, Cystoscopy,Laparoscopy etc. This work uses MagneticResonance Imaging (MRI) as the diagnosing tool.That is the cervical cancer MRI in dicom (DigitalImaging and Communications in Medicine) format  665SOUMYA et al. , Biomed. & Pharmacol. J., Vol. 9 (2), 663-671 (2016)is taken as the input to this work. Convert the imageinto gray scale and remove the noise and improvethe image quality to get more surety and ease indetecting the tumor. The dataset consist of magneticresonance images of 24 patients with locallyadvanced cervical cancer. The three datasets, axialT1and T2-weighted images and sagital T2-weightedimages are considered for this work. Algorithm Step 1Input the MR image.Step 2Enhance the MR image using Adaptivegamma correction method.Step 3The enhanced image is then segmentedusing Otsu’s segmentation technique.Step 4:Features are now extracted using 3methods:1.Grey level co-occurrence matrix (GLCM)features are extracted2.Contourlet transforms feature extraction3.Gabor filter feature extractionStep 5Classification of the image is done usingSupport vector machine (SVM) classifier.Step 6Staging of each image is the resultantoutput. Pre processing It is the first step in our proposedtechnique. Acquired images are then preprocessedusing the following methods. Image enhancement Enhancement improves the brightness ofdimmed images via the gamma correction andprobability distribution of luminance pixels. Theenhancement techniques are classified into two:direct enhancement methods and indirectenhancement methods. In direct enhancementmethods, the image contrast can be directly definedby a specific contrast term. In indirect enhancementmethods attempt to enhance image contrast byredistributing the probability density. The adaptivegamma correction (AGC) 9  is formulated as follows: (1)Gamma parameters calculated as:...(2)The image contrast can be enhancedrobotically by Gamma correction, which is a non-linear operation. Gamma correction is used forcorrecting lightness or the darkness of image[4].Image brightness can be corrected according tothe gamma value only. Range of Gamma value isfrom 0.0 to 10.0. If the gamma value is less than1.0(<1.0), then the image gets darken. Else if thegamma value is greater than 1.0(>1.0), the imagegets lighten. Else gamma value is equal to 1, thenno changes in an image. Gamma is applied onlyfor display images not to the data of image.The AGC method can progressivelyincrease the low intensity and avoid the significantdecrement of the high intensity. Three major processof gamma correction are:ãHistogram analysis – It provides the spatialinformation of an input image.ãWeighting distribution – It is used to eventhe irregular occurrence and thus avoidgeneration of inauspicious artifacts.ãGamma correction – It can roboticallyenhance the image. Image segmentation An important part in image processing isimage segmentation. The divisions of an image intomeaningful structures are called Imagesegmentation. The purpose of segmentation is to Fig. 1: Block diagram  666SOUMYA et al. , Biomed. & Pharmacol. J., Vol. 9 (2), 663-671 (2016) Fig. 3c: Histogram 1Fig. 3a: Input imageFig. 3b: Enhanced imageFig. 3d: Histogram 2Fig. 3f: Contourlet featureFig. 3e: Segmented Image change the representation of an image intosomething that is more meaningful and easier toanalyze. The methods are categorized on the basisof two properties discontinuity and similarity. Basedon this property image segmentation is categorizedas Edged based segmentation and region basedsegmentation. The segmentation methods that arebased on discontinuity property of pixels areconsidered as boundary or edges basedtechniques. The region based segmentation ispartitioning of an image into similar areas ofconnected pixels. There are different type of theRegion based method like thresholding, regiongrowing and region splitting and merging.Thresholding is a main technique in imagesegmentation applications. The basic idea is toselect an optimal gray-level threshold value fordividing objects of interest in an image from thebackground based on their gray-level distribution. Otsu method is an example of global thresholdingin which it depend only gray value of the image. It isproposed by Scholar Otsu in 1979. Otsu method iswidely used because it is simple and effective. TheOtsu method requires computing a gray levelhistogram before running 10-11 .
Similar documents
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