A Study of Brain Tumor MRI Image Using Distance Based Clustering Algorithms

Transcription

1 A Study of Bra Tuor MRI Iage Usg Dstace Based Clusterg Algorths [] Dr. O.A. Mohaed Jafar, [] J. Zehara Sulthaa [] Assocate Professor, [] Research Scholar [] Coputer Scece, Jaal Mohaed College (Autooous), Truchrappall. [] Jaal Mohaed College (Autooous), Truchrappall Abstract Bra tuor segetato s a challegg proble. The edcal expert ca aalyze the MRI (Magetc Resoace Iagg) but ths actvty s error-proe ad te cosug process. Data clusterg s the process of groupg slar obects. Exstg bra tuor segetato ethods use Eucldea dstace etrc. I ths paper, a study of hard ad fuzzy ethods based o dfferet dstace etrcs Mahatta ad Chebyshev used for bra tuor segetato s doe. The result of Adaptve Fuzzy K-eas (AFKM) s copared wth K-eas ad Fuzzy K-eas (FKM) algorths. The AFKM based o Chebyshev dstace gves better result tha K-eas ad FKM techques ters of accuracy ad quattatve easureets. Keywords Iage Segetato, Clusterg, K-eas, FKM, AFKM, MRI, Dstace Metrcs.. INTRODUCTION Hua bra s the portat orga whch cotrols all our physcal ad etal tasks. Bra tuor s a aboral or ucotrolled growth of cells. It ay crease pressure the bra ad ay daage the cetral ervous syste. Recetly, bra tuor detecto s oe of the sgfcat tasks. Bra tuor s broadly classfed as beg ad algat. A beg tuor s ot cacerous ad the tuor cells do ot spread to other parts of our body. Malgat tuor s cacerous ad the cells grow out of cotrol. It does ot have clear borders []. Radologsts exae the patets by usg varous agg techques such as X-rays, Ultra-soud, Coputerzed Toography (CT) sca, Postro Esso Toography (PET), ad Magetc Resoace Iagg (MRI). Medcal agg plays a key role to dagose the bra tuor ad helps radologsts for better treatet []. X-rays: They easure the desty of tssues. They are used to aalyze ad see boe dsease, boe breaks, fracture, degeerato, dslocatos ad fectos. Ultra-soud: It s used to fd the locatos of surfaces wth tssues ad dfferetates surfaces fro fluds. CT sca: It produces hgh-qualty, detaled ages of the body. It produces detaled ages of blood vessels, boes, orgas, ad soft tssue. It s ore powerful tha X-rays. PET: It easures the physologcal fuctog of a bra. It s aly used to study the expresso of specfc gees the bra, eurotrastters ad actos of pharaceutcal drugs. MRI sca: It provdes detaled forato about cellular structure ad bra tuor aatoy. It does ot cota ay radatos. It s a excellet tool used by doctors to detect aboralty bra. It s ore cofortable tha other scag ethods due to hgh tssue cotrast, better dfferetato of tssues ad has fewer artfacts. Iage segetato s the process of sub-dvdg a age to sets of pxels. It helps locatg the bra tuor boudary rego approprately. The segetato techques ca be categorzed to edge based segetato, threshold based segetato, rego based segetato ad cluster based segetato. I edge based segetato techques, the abrupt chage testy values s take to cocer for obect extracto. I threshold based segetato techques, the obects fro the age are extracted o the bass of a partcular threshold. Ths techque s approprate for ages wth dfferet testes of pxels. I rego based segetato techques, the age s dvded to regos that are slar testes o the bass of soe predefed crtero. I cluster based segetato techques, a age s dvded to a uber of clusters based o the value of ebershp fuctos allotted to each pxel the age [3] Clusterg s a ethod that attepts to partto a set of pxels to groups such that pxels each subset are slar to each other. Clusterg ethods are broadly classfed to hard ad soft clusterg. The hard clusterg techque, K-eas s sple ad wdely used ethod for age segetato [4]. But, t s ot suted for real world applcatos. The soft clusterg techque, Fuzzy K-eas s a effcet ethod. It perts us data eleets belog to ore tha oe cluster. Recetly All Rghts Reserved 08 IJERCSE 80

2 AFKM s foud success solvg age segetato probles. Dstace easure plays a key role clusterg aalyss. I ths paper, segetato va varous clusterg ethods s used to seget the MRI bra age based o dfferet dstace easures such as Eucldea, Mahatta ad Chebyshev The reader of the paper s orgazed as follows: I secto, related work s descrbed. The ethodology s explaed secto 3. Results ad dscusso are preseted secto 4. Fally, cocluso s gve secto 5. RELATED WORK Ths secto gves wth the works related to aalyss of bra tuor detecto MRI ages. Logeswar ad Kara [5] descrbed segetato ethod two phases. They have studed the perforace of the MRI age ters of weght vector, executo te ad tuor pxels detected. Vasuda ad Satheesh [6] developed proved fuzzy c-eas algorth for MR bra age segetato. I ther proposed ethod, the coputatoal rate s proved by odfyg the cluster ceter ad ebershp values. TaeSelvy et al. [7] aalyzed varous clusterg techques such as K-eas, SOM, Herarchcal Clusterg ad Fuzzy C-Meas to track tuor obects Magetc Resoace (MR) bra ages. Padole ad Chaudhar [8] revewed varous segetato ethods edcal ages. They have cobed the ea shft ad oralzed cut ethods ther ew approach. Patl ad Bhalchadra [9] proposed a strategy to detect ad extracto of bra tuour fro patet s MRI sca ages of the bra by usg MATLAB software. Vay ad Subhash [0] descrbe a effcet ethod for autoatc bra tuor segetato for the extracto of tuor tssues fro MR ages. They have proposed K-eas algorth for better perforace ad good results. Ala Jose et al. [] proposed segetato ethods for bra tuor usg K-eas ad Fuzzy C-eas algorths. Before the process of algorth, the ose preset the MR age s frst reoved. The ose free age s gve as put to the clusterg algorths for extractg tuors fro MR age. Vshal Shde et al. [] proposed a segetato ethod that uses the K- eas clusterg techque to detfy the tuor MRI. I ther work, K-eas ethod s used to covert a gve RGB age to gray scale age ad the separate the posto of tuor obects. The a of ther study s the desg of a coputer syste that able to detfy the presece of dgtal ages of the bra ad to accurately defe ts borderles. Sagar Al Kuar ad Chadra Sekhar Rao [3] pleeted K-eas based segetato ethod to detect bra tuor ad cacer cells. Ths segetato ethod ehaces the growth boudares ad fast wth other clusterg ethods. Ther algorth produces relable results wth less executo te ad less uber of teratos. Rabab Saadoo Abdoo et al. [4] pleeted the Fuzzy C-Mea clusterg to seget three aboral bra MR ages. The results dcate that, for hgh ebershp values, error creases due to the hgh correlato betwee the bra tssues. Sddh Nerurkar[5] preseted the two effcet age segetato algorths K-Meas ad Rego Growg Techques for bra MRI ages. Cacerous Tuors have bee detected fro the bra MRI scas usg these two techques. Rego Growg Techque s foud to be ore accurate tha K-Meas algorth. Gayathr advasath [6] preseted the K- eas ad fuzzy c-eas algorths for age segetato. The bra tuor s detected ad ts exact locato s detfed. METHODOLOGY Pre-processg Pre-processg s essetally for to correct ad adust the age for further study. Dfferet types of flterg techques are avalable the lterature. These flters are used to suppress the ose, soothe ad ehace the age ad prove the qualty the age. Soe portat flters are Meda flter, Adaptve Meda flter, Mea flter, Adaptve ea flter, Gaussa Soothg flter ad Weer flter. These flters are used for MRI bra age pre-processg. The eda flter s oe of the ost faous flters whch s sple to pleet. It s a olear dgtal flterg techque, used to reove ose fro a age. Such ose reducto s a typcal pre-processg step to prove the results of later processg. It s a good techque to reove salt ad pepper type of ose. It works by ovg through the age pxel by pxel, replacg each value wth the eda value of eghborg pxels. [7] Dstace Metrcs Dstace easure plays a portat role clusterg. It s essetal to detfy the aer whch the data eleets are related ad how the data eleets are slar ad dsslar wth each other. The portat propertes All Rghts Reserved 08 IJERCSE 8

3 of dstace etrcs are o-egatvty, syetry, detfcato, defteess ad tragle equalty [8]. If the pots have deso such as p ( x, x,... x ) ad q ( y, y,... y ) the varous dstace etrcs are calculated fro the followg table: Dstace Metrc Eucldea Mahatta Forula d(p,q) ( x y ) x y Chybyshev Max,,... x y Clusterg Techques The cluster aalyss s a optzato proble whch a gve fucto cosstg of tra-cluster slarty ad ter-cluster dsslarty eeds to be optzed. It has a wde rage of applcatos ay dfferet areas lke arketg research, age aalyss, age segetato, edce, character recogto, bakg ad face web aalyss, coputer forescs etc. K-eas Clusterg Algorth K-eas clusterg techque s oe of the splest hard clusterg techques. Gve uber of data eleets, ths algorth groups these data to specfed clusters. The data eleets belog to the sae cluster are slar ature ad those belogg to dfferet clusters are dfferet ature [9] [0]. X {x,x...x } Let c { c, c... c } k be the set of data pots ad be the set of cluster ceters. Step: Frst select k cluster ceters radoly. Step : Calculate the dstace betwee the pxels ad cluster ceters. Choose the sutable dstace fucto. Step 3: Assg each pxel x to ts earest cluster ceter c Step 4: Calculate the ew cluster ceter c usg c x x where to the -th cluster s the uber of pxels belogg Step 5: Repeat the steps () to (4) utl covergece s reached. Fuzzy K-eas Clusterg Algorth Fuzzy K-eas (FKM) or Fuzzy c-eas (FCM) algorth s a popular data clusterg techque for aalyss of MRI ages []. It s a exteso of K- eas algorth. FKM s a clusterg algorth whch each data pot belogs to cluster to a degree specfed by a ebershp grade. The pxels ear to the cluster ceter have hgh ebershp degree towards the partcular cluster ceter []. Paraeters: u - fuzzy partto ebershp values; x - data values; z - th cluster ceter; fuzzess expoet; - terato error Step : Italze Step : Detere z u x u (0) U z,,,... c, usg Step 3: Detere the sutabledstace easure Step 4: Calculate the obectve fucto value J c u d Step 5: Update ( k ) U usg u ; c ( k ) c k ( k) ( k) d d k ; J usg U U the stop; otherwse go Step 6: If to step Adaptve Fuzzy K-eas Clusterg Algorth AFKM s proposed by Sulaa et al. Ths ethod s wdely used age segetato [3] [4][5]. I ths paper, AFKM clusterg algorth wth ew dstace easures s proposed. The obectve fucto value of AKFM s calculated usg the equato All Rghts Reserved 08 IJERCSE 8

4 c J ( M ) d where k t M tk s the fuzzy ebershp fucto ad s the fuzzess expoet The cluster ceter s detered usg the equato C k where N t N t defed as ' ( M ) v ' ( M ) ' ( M ) ( M ) M M t s the ew ebershp fucto ad s ' M ( c )( e ) k k ad e k s the error of beloggess. RESULTS AND DISCUSSION I ths paper, the followg MRI bra age take fro data base s selected for aalyss by the clusterg algorths based o dfferet dstace easures. The orgal age of MRI bra ages s show fg.. The results are aalyzed based o accuracy, F(I) ad ea squared error (MSE) [6] [7]. The varous perforace easures are descrbed as follows: TN TP Accuracy TN TP FN FP where TN s true postve, TP s true postve, FN s false egatve ad FP s false postve. R F() I R e N s A M MSE x c Fg. Orgal Iage of MRI Bra Iages Tables I to VI ad Fg. to 7 show the results of varous perforace easures for the selected age. Fro the results, the AFKM clusterg ethod based o Chebyshev dstace easure produces better result tha the other clusterg ethods, aely K-Meas ad FKM for dfferet etrcs. Table I: Evaluato of Accuracy for Iage Algorth Dstace Metrc Eucldea Mahatta Chebyshev K-eas FKM AFKM Table II: Evaluato of Accuracy for Iage Algorth Dstace Metrc Eucldea Mahatta Chebyshev K-eas FKM AFKM Table III: Evaluato of F(I) for Iage Algorth Dstace Metrc Eucldea Mahatta Chebyshev K-eas FKM AFKM Table IV: Evaluato of F(I) for Iage Algorth Dstace Metrc Eucldea Mahatta Chebyshev K-eas FKM AFKM Table V: Evaluato of MSE for Iage Algorth Dstace Metrc Eucldea Mahatta Chebyshev K-eas FKM AFKM (a) (b) All Rghts Reserved 08 IJERCSE 83

5 Table V: Evaluato of MSE for Iage Algorth Dstace Metrc Eucldea Mahatta Chebyshev K-eas FKM AFKM Fg. 5 Evaluato of F(I) for Iage Fg. Evaluato of Accuracy for Iage Fg. 6 Evaluato of MSE for Iage Fg. 3 Evaluato of Accuracy for Iage Fg. 7 Evaluato of MSE for Iage Fg. 4 Evaluato of F(I) for Iage All Rghts Reserved 08 IJERCSE 84

6 CONCLUSION Bra tuors are caused by aboral growth of cells sder the bra. Bra tuor segetato s oe of the portat edcal dagostc techques. The perforace of bra tuor segetato s evaluated usg K-eas, FKM ad AFKM Clusterg Algorths wth three dstace easures such as Eucldea, Mahatta ad Chebyshev dstace etrcs based o MRI Iages. Fro the coparso, t s observed that the AFKM Clusterg Algorth usg Chebyshev dstace gves better results tha K-eas ad FKM ters of Accuracy, ad Quattatve Measureets. REFERENCES [] Kothar, A Study o Classfcato ad Detecto of Bra Tuor Techques, Iteratoal Joural of Coputer Egeerg ad Techology, Vol. 6 (), pp , 05. [] Deepa, Akasha Sgh, Revew of Bra Tuor Detecto fro MRI Iages, IEEE 06, pp [3] K.S. Agel V, J. Jayakuar, Perforace evaluato of stadard age segetato ethods ad clusterg algorths for segetato of MRI bra tuor ages, Europea Joural of Scetfc Research, Vol. 79, No., pp , 0. [4] J. Vay, J. Subhash, A Effcet Bra Tuor Detecto Methodology Usg K-Meas Clusterg Algorth, IEEE Iteratoal Coferece o Coucato ad Sgal Processg, pp , Aprl 3-5, 03. [5] T. Logeswar ad K. Kara, A proved pleetato of bra tuor detecto usg segetato based o soft coputg, Joural of Cacer Research ad Experetal Ocology, Vol., Issue, pp , 00. [6] P. Vasuda ad S. Satheesh, Iproved Fuzzy C- Meas Algorth for MR Bra Iage Segetato, Iteratoal Joural o Coputer Scece ad Egeerg (IJCSE), Vol. 0, No. 5, pp , 00. [7] P. TaeSelvy, V. Palasay ad T. Purusothaa, Perforace Aalyss of Clusterg Algorths Bra Tuor Detecto of MR Iages, Europea Joural of Scetfc Research, Vol. 63, No. 3, pp , 0. [8] Vshal B. Padole ad D.S. Chaudhar, A Revew of Segetato Methods for Detecto of Bra Tuor MRI, Iteratoal Joural of Electrocs, Coucato & Soft Coputg Scece ad Egeerg (IJECSCSE), Vol., Issue, pp. 5-8, 0. [9] Raesh C. Patl ad Dr. A.S. Bhalchadra, Bra Tuour Extracto fro MRI Iages Usg MATLAB, Iteratoal Joural of Electrocs, Coucato & Soft Coputg Scece ad Egeerg (IJECSCSE), Vol., No., pp. -4, 0. [0] J. Vay ad J. Subhash, A Effcet Bra Tuor Detecto Methodology Usg K-Meas Clusterg Algorth, IEEE Iteratoal Coferece o Coucato ad Sgal Processg, pp , Aprl 3-5, 03. [] Ala Jose, S. Rav ad M. Sabath, Bra Tuor Segetato Usg K-Meas Clusterg ad Fuzzy C-Meas Algorths ad ts Area Calculato, Iteratoal Joural of Iovatve Research Coputer ad Coucato Egeerg, Vol., Issue 3, pp , 04. [] Vshal Shde, PrtKe, SuchtaGadge ad ShekharKhatal, Bra Tuor Idetfcato usg MRI Iages, Iteratoal Joural o Recet ad Iovato Treds Coputg ad Coucato, Vol., Issue 0, pp , 04. [3] V. Sagar Al Kuar ad T. Chadra Sekhar Rao, Bra Tuor Extracto by K-eas Clusterg Based o Morphologcal Iage Processg, Iteratoal Joural of Egeerg ad Coputer Scece, Vol. 3, Issue 0, pp , 04. [4] Rabab Saadoo Abdoo, Loay Kado Abood ad S.M. Al, Aalyss Study of Fuzzy C-Mea Algorth Ipleeted o Aboral MR Bra Iages, Vol. 7, Issue 3, pp. 4-49, 05. [5] Sddh N. Nerurkar, Bra Tuor Detecto usg Iage Segetato, Iteratoal Joural of Egeerg Research Coputer Scece ad Egeerg, Vol. 4, Issue 4, pp , 07. [6] K. Gayathr ad D. Vasath, Bra Tuor Segetato Usg K-Meas Clusterg ad Fuzzy C-Meas Algorths, Iteratoal Joural of Scetfc Research Coputer Scece, Egeerg ad Iforato Techology, Vol., Issue, pp , 07. [7] Azadeh Noor Hoshyar, Adel Al-Jualy ad Afsaeh Noor Hoshyar, Coparg the Perforace of Varous Flters o Sk Cacer Iages, Proceda Coputer Scece, 4, pp. 3-37, 04. All Rghts Reserved 08 IJERCSE 85

7 [8] Elea Deza ad Mchel Mare Deza, Ecyclopeda of Dstaces, Sprger, 009. [9] J. MacQuee, Soe ethods for classfcato ad aalyss of ultvarate observatos, Proceedgs of the Ffth Berkeley Syposu o Matheatcal Statstcs ad Probablty, Vol., pp. 8-97, 967. [0] H. Kasha, J. Hu, B. Ray ad M. Sgh, K-eas clusterg of proportoal data usg L dstace, Proceedgs of the 9 th Iteratoal Coferece o Patter Recogto, pp. -4, Dec. 8-, 008. [] YE Pg, Fuzzy K-eas algorths based o ebershp fucto proveet, Chagchu Isttute of Techology (Natural Sceces Edto), 007. [] J.C. Bezdek, Patter recogto wth fuzzy obectve fucto algorths, Pleu Press, New York, 98. [3] S.N. Sulaa ad N.A.M. Isa, Adaptve fuzzy-keas clusterg algorth for age segetato, IEEE Trasactos o Cosuer Electrocs, Vol. 56, pp , 00. [4] St Nora Sulaa, Norela Awag No, Iza Sazata Isa, Norhaz Hazah, Seetato of bra MRI Iage based o clusterg algorth,, Recet Advaces Electrcal ad Coputer Egeerg, pp. 36-4, 04. [5] N. Jadhav Swapl ad Prof. Sarta V. Vera, Perceptble Perforace of Dfferet Clusterg Techques for Iage Segetato, Iteratoal Joural of Scetfc Research Egeerg & Techology (IJSRET), Vol. 3, Issue 5, pp , 04. [6] Y.H. Yag ad J. Lu, Multresoluto Color Iage Segetato, IEEE Trasactos o Patter Aalyss ad Mache Itellgece, Vol. 6, pp , 994. [7] Zhou Wag, Ala Corad Bovk, Had Rah Shekh, Eero P. Socell, Iage Qualty Assesset: Fro Vsblty to Structural Slarty, IEEE Trasactos o Iage Processg, Vol. 3, No. 4, pp , 004. All Rghts Reserved 08 IJERCSE 86