BRAIN MRI SEGMENTATION AND TUMOR DETECTION USING FCM AND NEURAL NETWORKS

Transcription

1 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May 2015 BRAIN MRI SEGMENTATION AND TUMOR DETECTION USING FCM AND NEURAL NETWORKS NIVEDHA.R PG SCHOLAR,DEPT OF ECE Mrs.ANGAYARKANNI.N, M.E,(Ph.D.) PROFESSOR/ECE PGP COLLEGE OF ENGINEERING AND TECHNOLOGY, NAMAKKAL nivedha1692@gmail.cm Abstract-The gal f ur prject is t present a nvel algrithm fr tumr segmentatin f magnetic resnance imaging (MRI) data f brain.ne f the mst cmmnly used methds fr Magnetic Resnance Imaging (MRI) segmentatin is Fuzzy C Means (FCM). This methd in cmparisn with ther methds Preserves mre infrmatin f the images.because f using the intensity f pixels as a key feature fr clustering, Standard FCM is sensitive t nise.in this prject in additin t intensity, mean f neighbrhd f pixels and largest singular value f neighbrhd f pixels are used as features. Als a methd fr tumr segmenting MRI images is presented which uses bth FCM feed frward neural netwrk and partly decreases the limitatin f standard FCMthe different stages f the tumr is als detected based n the area cverage during tumr segmentatin by feed frward neural netwrks Keywrds-tumrs Segmentatin,Magnetic Resnanceimage(MRI),Fuzzy-Means(FCM), feed frward neural netwrks(ffan) I.NTRODUCTION Tday because f increasing the number f medical images using cmputer fr analyzing and prcessing is seemed t be necessary. Specially using cmputerized algrithm fr autmatin f diagnsis and as a specialist helper is very imprtant. These algrithms that are named image segmentatin algrithms are used in many applicatin f medical imaging. One f the mst cmmnly used methds fr MRI Segmentatin is dne using FCM. FCM is an unsupervised methd. Since FCM des nt use lcal infrmatin f pixels, it is very sensitive t nise Medical images due t different reasns are affected by nise; therefre segmenting them t bvius segmented clusters is difficult. When a pixel is affected by nise, intensity f the pixel is changed. In this situatin neighbrhd infrmatin shuld be used. T vercme this drawback, in ur in additin t intensity, neighbrhd infrmatin is used as input features in FCM and FEED FORWARD NEURAL NETWORKS A tumr may be primary r secndary. If it is the rigin, then it is knwn as primary. If the part f the tumr spreads t anther place andgrws n its wn, then it is knwn as secndary. The brain tumr affects CSF (Cerebral Spinal Fluid)and causes strkes. The physician gives the treatment fr the strkesrather than the treatment fr tumrs. S the detectin f the tumr isimprtant fr that treatment, s then ur prject enables t determine the stage f the tumr Autmated brain tumr detectin frm MRI images is ne f the mst challenging task in tday s mdern Medical imaging research. Magnetic Resnance Images are used t prduce images f tumr in human brain. It is used t analyze the human brain withut the need fr surgery 1.1Objective The bjective f ur prject is t intrduce segmentatin f MRI image f the human brain. In additin with the help f FCM segmentatin and 77

2 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May 2015 FEED FORWARD NEURAL NETWORKS we may determine the different stages f brain tumr 1.2Overview f prject Our prject is t implement f Simple Algrithm fr detectin f range and shape f tumr in brain MR Images. Nrmally the anatmy f the Brain can be viewed by the MRI scan r CT scan. MRI scanned image is used fr the entire prcess. The MRI scan is mre cmfrtable than any ther scans fr diagnsis. It will nt affect the human bdy, because it desn t practice any radiatin. It is centered n the magnetic field and radi waves. There are dissimilar types f algrithm were develped fr brain tumr detectin. But theymay have sme drawback in detectin and extractin The segmentatin, which is dne thrugh fuzzy c-means The life expectancy f the persn wh is affected by the brain tumr will increase if it is detected at an earlier stage. Nrmally tumr cells are f tw types Mass and Malignant. The detectin f the malignant tumr is smewhat difficult t mass tumr. Our prject fcuses n the detectin f any stage f brain tumr. The develpment platfrm fr thedetectin is matlab because it is easy t develp and execute. At the end, we are prviding systems that detect the tumr and its shape and helps fr further treatment II. PROPOSED METHOD 2.1 Intrductin The techniques used fr this survey are Brain Tumr Detectin Using Segmentatin Fuzzy C- MeansTechnique with NEURAL NETWORK Initializatin fr Brain Tissue Segmentatin fr tumr detectin One f the limitatins f FCM is that it desn t cnsider lcal infrmatin f pixels. In rder t vercme this limitatin n additin f intensity f pixel, mean f 3 3 neighbrhd f pixels and largest singular value f 5 5 neighbrhd f pixels are used as features 2.2Advantages The purpse f image segmentatin is tpartitin an image int meaningful regins with respect t a particular applicatin The segmentatin is based n measurements taken frm the image and might be grey level, clur, texture, depth r mtin Usually image segmentatin is an initial and vital step in a series f prcesses aimed at verall image understanding Enables bject surfaces with varying patterns f grey t be segmented 2.3 Applicatins Identifying bjects in a scene fr bject-based measurements such as size and shape Identifying bjects in a mving scene fr bject-based vide cmpressin (MPEG4) Identifying bjects which are at different distances frm a sensr using depth measurements frm a laser range finder enabling path planning fr a mbile rbts 3.FUNDAMENTALDEFINITIONAND DESCRIPTIONS 3.1 Fuzzy c-means clustering FCM partitins a set f n bjects x {x,x,..., xn} in Rd dimensinal space int c(1 <c <n) fuzzy clusters with y = {y1, y2, y3,..., yc} cluster centers r centrids [6]. The fuzzy clustering f bjects is described by a fuzzy matrix μ with n rws and c clumns in which n is the number f data bjects 78

3 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May 2015 and c in the number f clusters. μij, the element in the ith rw and jth clumn in μ, indicates the degree f assciatin r membership functin f the ith bject with the jth cluster. The bjective functin f FCM algrithm is t minimize the fllwing equatin Where nuclei deep inside f the brain, including the thalami and basal ganglia. The WM cnstitutes a cnnected regin that is brdered by GM and CSF the display purpse WM is shwn in gray clr, GM as white clr and CSF as black clr. In MRI f head scans, the picture f rgan isusually surrunded by air particles, knwn as backgrund (bck) in rder t make a matrix representatin. This bck is anther majr ROI in MRI f head scan In this survey there are varius latest techniques used in image segmentatin which are very useful in medical field fr diagnsis f a prblem. m(m > 1) is a scalar termed as weighting expnent. M cntrls the fuzziness f the resulting clusters and dijis the Euclidian distance frm bject i x t the cluster center y j. The yj, centrid f the jth cluster, is btained as: The FCM algrithm is iterative and can be stated as fllws: 1. Select m(m > 1); initialize the membership functin values μij, i= 1,2,..., n; j = 1,2,...,c. 2. Cmpute the cluster centers yj, j = 1,2,...,,c.accrding t equatin (3) 3. Cmpute Euclidian distance dij,i=1,2,n j=1,2,,c. 3.2 characteristics f mri scans The images prduced by MRI scans are usually gray images with intensity in the range The GM f the brain cnsists f the crtex that lines the external surface f the brain and the gray Fig1.Segmentatin results f MRI Lw depth f Field methd is a methd used t give special imprtance t a part f image which is essential r which has t be fcused. This methd can be used in the fields like sprts, phtgraphy & medical. In medical field prblem diagnsis can be very easy by making fcus t a prblematic area such as Brain tumr, Cancer etc. the area which is fcused i.e. Object f Interest (OOI) helpful t prvide new infrmatin & help in research as well. Here, image segmentatin algrithm is useful t separate Object f Interest (OOI) frm the rest f the image. The algrithm fr develping such type f images divided int five parts Deviatin Scring, Scre Clustering, Mask Apprximatin, Clr Segmentatin and Regin Scring. In the initial stage f the algrithm i.e. Deviatin scring the Gaussian Blur algrithm is used. Accrding t this the image divided in tw ways ie. Original image & surface cut f OOI & the difference is calculated frm these images which gives a scre value. In the secnd stage f the algrithm. 79

4 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May IV.FLOWDIAGRAM pixels) will be cmpared against the ne that is tw pixels t the right. Measure Texture quantizes the image int eight intensity levels. There are then 8x8 pssible ways t categrize a pixel with its scaleneighbr. Measure Texture frms the 8x8 cccurrence matrix by cunting hw many pixels andneighbrs have each f the 8x8 intensity cmbinatins. Angular Secnd Mment Cntrast Crrelatin Variance InverseDifferenceMment Sum Average Sum Variance 4.2 Image texture An image texture is a set f metrics calculated in image prcessing designed t quantify the perceived texture f an image. Image Texture gives us infrmatin abut the spatial arrangement f clr r intensities in an image r selected regin f an image. [1] 4.1Different Texture Features algrithms: Haralick features GLCM (gray level c-ccurrence matrix) Gabr filter Haralick Features:Haralick texture features are derived frm the c-ccurrence matrix, which cntains infrmatin abut hw image intensities in pixels with a certain psitin in relatin t each ther ccur tgether. Measure Texture can measure textures at different scales; the scale yu chse determines hw the c-ccurrence matrix is cnstructed. Fr example, if yu chse a scale f 2, each pixel in the image (excluding sme brder Image textures can be artificially created r fund in natural scenes captured in an image. Image textures are ne way that can be used t help in Segmentati r classificatin f images. T analyze an image texture in cmputer graphics, there are tw ways t apprach the issue: Structured Apprach and Statistical Apprach V.TUMOR SEGMENTATION Digital image prcessing is vast fields which can be using varius applicatins. Which include Detectin f criminal face, figure print authenticatin system, in medical field, bject recgnitin etc. Brain tumr detectin plays an Imprtant rle in medical field. Brain tumr detectin is detectin f tumr affected part in the brain alng with its shape size and bundary, s it useful in medical field 80

5 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May Types f tumr Tumr The wrd tumr is a synnym fr a wrd neplasm which is frmed by an abnrmal grwth f cells Tumr is smething ttally different frm cancer. There are three cmmn types f tumr: 1) Benign 2) Pre-Malignant 3) Malignant Benign Tumr A benign tumr is a tumr is the ne that des nt expand in an abrupt way; it desn t affect its neighbring healthy tissues and als des nt expand t nn-adjacent tissues. Mles are the cmmn example f benign tumrs. Pre-Malignant Tumr Premalignant Tumr is a precancerus stage, cnsidered as a disease, if nt prperly treated it may lead t cancer. Malignant Tumr Malignancy (mal- = "bad" and -ignis = "fire") is the type f tumr, that grws wrse with the passage f time and ultimately results in the death f a persn. Malignant is basically a medical term that describes a severe prgressing disease. Malignant tumr is a term which is typically used fr the descriptin f cancer. VI.TUMOR SEGMENTATION ANDDETECTION ALGORITHM step1:initially the human brain image is btained frm magnetic resnance image(mri) step2: Nw the brain tissues are segmented using fuzzy c mean algrithm(fcm) step3: Based n the FCM algrithm the intensity f each segmented image is calculated step4:calculate Intensity; mean f 3 3 neighbrhd fpixels and largest singular value f 5 5 neighbrhdf pixels fr each pixel. step5: Nw the intensity mean and the largest singular value is calculated s then the brain tumr can be segmented with high intensity clured image step6: Nw the tumr is segmented; frm the segmented clured pixel image the cverage area f the tumr is calculated by feed frward neural netwrks step7: Based n the cverage area f the tumr we may determine whether the tumr is in initial, middle r final stage step8: Finally the brain tumr is segmented and the tumr stage is detected successfully VII.TUMOR DETECTION 7.1Image prcessing techniques As has just been established, a number f factrs can adversely affect RTR image quality. With the use f image enhancement techniques, the difference in sensitivity between film and RTR can be decreased. A number f image prcessing techniques, in additin t enhancement techniques, can be applied t imprve the data usefulness. Techniques include cnvlutin edge detectin, mathematics, filters, trend remval, and image analysis. The varius image enhancements and image prcessing techniques will be intrduced in this sectin. Cmputer sftware prgrams are available, including sme r all f the fllwing prgrams: Enhancement prgrams make infrmatin mre visible. Histgram equalizatin-redistributes the intensities f the image f the entire range f pssible intensities (usually 256 grayscale levels). Unsharp masking-subtracts smthed image frm the riginal image t emphasize intensity changes. 81

6 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May Image analysis Gray-scale mapping-alters mapping f intensity f pixels in file t intensity displayed n a cmputer screen. Slice-Plts intensity versus psitin fr hrizntal, vertical, r arbitrary directin. Lists intensity versus pixel lcatin frm any pint alng the slice. Image extractin-extracts a prtin r all f an image and creates a new image with the selected area. 3 TUMORIDENTIFICATION FIG VIII.SIMULATED RESULTS FIG 4 BENIGN TUMOR FIG 2 NORMAL MRI IMAGE 82

7 Internatinal Jurnal f Advanced Research in Bilgy, Eclgy, Science and Technlgy (IJARBEST) Vl. 1, Issue 2, May 2015 FIG 5 MALIGNANT TUMOR VIII. CONCLUSION Thus the brain MRI segmentatin can be dne using fuzzy c means and with the help f neural netwrks we may detect the stage f the tumr.there are different types f tumrs available. It may benign r malignant ver the brain. Segmentatin is dne using fuzzy c means algrithm and neural netwrks detects the tumrstage. Achieved results are shwn in upper sectin which shws the efficient tumr detectin by using FCM algrithm rather than any ther algrithm and als finding the bundary extractin f tumr.the stage f the tumr is based n the area f tumr. Shape and Size f the tumr and the stage f the tumr can be detected successfully based n new kernel-induced distancemeasure, IEEE Trans. Systems Man Cybernet. B 34 (4) 2013 [4] Alan Jse, S.Ravi, M.Sambath, Brain Tumr Segmentatin Using K-Means Clustering And Fuzzy C-Means Algrithms And Its Area Calculatin, Vl. 2, Issue 3, March 2014 [5] Shan Shen, William Sandham, Member, IEEE, Malclm Granat, and Annette Sterr, MRI Fuzzy Segmentatin f Brain Tissue Using Neighbrhd Attractin With Neural-Netwrk Optimizatin, VOL. 9, NO. 3, SEPTEMBER 2013 [6] S.JaveedHussain. SatyaSavithri, P.V. Sree Devi, Segmentatin f Tissues in Brain MRI Images using Dynamic Neur-Fuzzy Technique,Vlume- 1, Issue-6, January 2012 [7]Ritu Agrawal, Manisha Sharma, Review f Segmentatin Methds fr Brain Tissue with Magnetic Resnance Images,vl [8] M. Murugeswari, and M.Gayathri, Tumr detectininmri brain image segmentatinusingphase cngruency mdified fuzzy c mean algrithm,vl. 1 Issue 2, April 2014 [9] Ivana Desptvić, Student Member,IEEE, EwutVansteenkiste, and Wilfried Philips, Member, IEEE, Spatially Cherent Fuzzy Clustering fr Accurate and Nise-Rbust Image Segmentatin, IEEE Transactins On Image Prcessing, Vl. 22, N. 2, APRIL IX.REFERENCE [1] D. Drmnt, D. Seidenwurm, D. Galanaud, P. Crnu, J. Yelnik, ande. Bardinet, Neurimaging and deep brain stimulatin, Amer. J. Neurradil.,vl. 31, pp , Jan [2]Reza Ghaderi,MehdiEzhi,and Jamal Ghasemi, Brain MRI Segmentatin f FCM and RBF neural netwrks,vl,31.pp.13,feb 2013 [3] S.C. Chen, D.Q. Zhang, Rbust image segmentatin using FCMwith spatial cnstraints 83