An Efficient System for Medical Image Retrieval using Generalized Gamma Distribution

Transcription

1 I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58 Published Online May 05 in MECS ( DOI: 0.585/ijigsp An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution T.V. Madhusudhana Rao Department of CSE, T.P.I.S.T., Bobbili, A.P., INDIA Dr. S.Pallam Setty Department of CS & SE, Andhra University, Visakhapatnam, INDIA Dr. Y.Srinivas Department of IT, GITAM University, Visakhapatnam, INDIA Abstrat Effiient diagnosis plays a ruial role for treatment. In many ases of ritialness, radiologists, dotors prefer to the usage of internet tehnologies in order to searh for similar ases. Aordingly in this paper an effetive mehanism of Content Based Image Retrieval (CBIR) is presented, whih helps the radiologists/dotors in retrieving similar from the medial dataset. The paper is presented by onsidering brain medial from a medial dataset. Feature vetors are to be extrated effiiently so as to retrieve the of interest. In this paper a two-way approah is adopted to retrieve the of relevane from the dataset. In the first step the Probability Density Funtions (PDF) are extrated and in the seond step the relevant are extrated using orrelation oeffiient. The auray of the model is tested on a database onsisting of 000 MR related to brain. The effetiveness of the model is tested using Preision, Reall, Error rate and Retrieval effiieny. The performane of the proposed model is ompared to Gaussian Mixture Model (GMM) using quality metris suh as Maximum distane, Mean Squared Error, Signal to Noise Ratio and Jaard quotient. Index Terms Generalized Gamma Distribution, Content Based Image Retrieval, Relevany, Correlation, Quality metris I. INTRODUCTION Brain disease is one of the most striking fators for the inrease in mortality rates. In India this rates have been inreasing ontinuously [4]. Most of the diseases ausing brain tumors may be either Benign or Malignant. The lassifiations of these are subjeted to the size of the tumors. Hene effetive diagnosis and identifiation of these diseases will be of ruial importane. Most of the ases related the treatment of these diseases is subjeted to MR imaging. MR imaging is mainly hoosing beause of its property of non-ionization. Along with the advantage, the grater disadvantage with these tehniques is that the reports are generated by the radiologists using the visual pereption of the MR []. However as the number of ases of brain related diseases are inreasing, the visual pereption may lead to inorret deisions. Another added fator to this is the proportionate inrease of the radiologists versus the inrease in the number of ases is lagging. Hene the advantage of using automated systems for effetive brain image sanning and report generations are of ruial importane. This paper highlights a methodology whih helps to retrieve the relevant from the databases. This system may help in partiular to the paramedis, radiologist residing in remote areas to suggest a basi treatment for patient in residing in rural areas. So that he/she an be shifted to nearby super speialized hospitals. Many systems have been presented in the literature based on Piture Arhive and Communiation Systems (PACS) [8]. Also text based retrieval systems are listed in the literature wherein the relevant text/keywords from the lab reports are retrieved against a query. However these systems have their own drawbaks sine text annotation is diffiult [4], though, the usage of text will be useful to some of the patients eager to know about some related fats about their disease. Hene, in this paper we present a novel methodology of CBIR using Generalized Gamma Distribution whih helps to retrieve similar based on ontent as well as text. The updated equations of the model parameters are estimated using Expetation Maximization (EM) algorithm. The main advantage behind the usage of this distribution is of the fat that the human tissues are asymmetri in nature and it an handle the spekles more effiiently. Assuming that the shape and texture features play a vital role in identifying the ontents more effetively [0][], in this model identifiation of Malignant and non-malignant tissues is experimented. Other models based on auto regression [7], wavelets [9], semanti gap [0][3], relevane feedbak [3], Gabor filter [4], - Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58

2 An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution 53 means algorithm [3][6], Geneti algorithm [9], D-Tree [8], binary splitting [], quad trees [] have been proposed in the literature. However, these models lak in effiieny while retrieving the of relevane beause of several disadvantages like -means algorithm is sensitive to initial lusters whih may not be effiient for handling medial where the data is both ontinuous and disrete. Models like binary splitting, D-Tree, Geneti algorithms are subjeted to parametri dependene and omplexity [5][5]. Hene effiient methods to retrieve the of interest will be of a great advantage. A. The Related Work Models based on Gamma, Log-Normal and Nakagami distributions are also highlighted in the literature for handling asymmetri distributions. However, Nakagami distribution overrules the other distributions while haraterizing the spekles in tissues [7]. Nevertheless the Nakagami distribution fails in handling larger impulse response of the spekles generated by human spekles ompared to Generalized Gamma Distribution. This paper presents a methodology using Generalized Gama Distribution. The organizing struture of the paper is as follows. The proposed methodology is presented in Setion II. The experimentation, results together with performane analysis and evaluation are highlighted in Setion III, the final Setion IV onludes the paper. II. PROPOSED METHODOLOGY USING GENERALIZED GAMMA DISTRIBUTION The proposed methodology is depited in the blok diagram shown in Figure, and it will be very muh useful for medial image analysis for ertain remote areas, where only the minimum first aid is available. Using the developed methodology, proedures an be evaluated and neessary treatments an be derived. These methods are deployed mainly on the brain image data olleted from University of Rennes database. Every brain image onsists of vital information suh as white matter (WM), Gray Matter (GM) and Cerebro Spinal Fluid (CSF). In order to assess the damage of the brain tissues, effetive methodologies are to be developed to disriminate WM, GM, and CSF. Fig.. Blok diagram of the proposed model A. Generalized Gamma Distribution Generalized Gamma Distribution is proposed for the lassifiation of the in this paper. In general, in Content Based Medial Image Retrieval, the damaged tissues are to be reognized. These damaged pixels behave in a different manner ompared to the nondamaged pixels and the intensity levels of these pixels will be more i.e. the number of non-damaged pixels are more than the damaged pixels. And these damaged pixels will have their peeks at the origin and hene the damaged medial will have their distributions with long tails. In order to interpret these distributions, tail distributions will be more appropriate and to ater these distributions the best suited distribution is Generalized Gamma Distribution. Generalized Gamma Distribution inludes several other distributions suh as Weibull distribution, Gamma distribution, Log-normal distribution, Chi-square distribution and Exponential distribution as partiular ases. The main advantage of using Generalized Gamma Distribution is that, it an sensitize the medial image features in presene of noise and with minimum variane more effetively. The Probability Density Funtion of the Generalized Gamma Distribution is of the form k ( x a) e f ( x, k,, a, b) k b ( k) xa b Where a, b and x are alled gamma variants and and k are alled shape parameters. By varying the value of the shape parameters, the partiular ases of gamma distribution an be modeled. B. Expetation Maximization (EM) Algorithm for Generalized Gamma Distribution In order to retrieve the relevant more effiiently and effetively, the parameters of the medial are to be estimated effetively. The parameters, Expetation Maximization algorithm is used for obtaining the final () Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58

3 54 An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution estimates. The final updates derived in [] and are presented here. N ( l) ( l) E [ t ( Z, )] () ( l) k N s x a k log b b k x x a a b b k s x a x a log b (3) a (4) b b k k x a t e 0 b x a k log e b log t t k dt k C. Image Database And Features Considerd In order to test the validity of the a database onsists of 000 pertaining to brain tumor disease is used. The olleted inlude, the inhomogeneities of the brain tumor suh as SEIZURES, MS-LESIONS and SCELEROSIS. Eah image is of fixed size whih is normalized and pre-proessed. In order to have a effetive reognition of the deformities, the features are to be extrated basing on position, olor and texture, sine any deformity in the brain image to be retrieved effetively needs to figure out the exat loation for whih the position of the (x, y) oordinates plae a vital role where x denotes the width and y denotes the height. The in-homogeneities in the brain an be identified and differentiate appropriately basing on the olor also. Hene for eah image the olor omponents are also onsidered. Another feature whih is of prime important for any retrieval and identifiation is the texture. In this methodology the texture of are extrated and these features are given as input to the Generalized Gamma Distribution. D. Image Mathing Using Correlation In this paper we present a novel methodology for identifying the relevany between the in the database and feature or a query image is onsidered for relevane. The main advantage of this methodology [] is that it an be used to test whether a subpart of the image (5) (6) belongs to the image in the database or not and it is given by C L, i (7) sum( L L ) sum n L Where L is the image from the database and is the query image (Note: and L should be of same dimension) and we say that the query image is the part of the image L, if the orrelation approahes to near. E. CBIR Algorithm For The Proposed Methodology Step : Obtain histogram of all the in the database, in order to model the various partiular ases of Generalized Gamma distribution Step : Calulate PDFs of Generalized Gamma Distribution Step 3: Find the relevant image based on orrelation using the formula C L, i (8) sum( L L ) sum n L Using the orrelation value, most relevant an be extrated. Step 4: Retrieve similar based on orrelation value of step 3 III. EXPERIMENTATION AND PERFORMANCE EVALUATION The brain from the image dataset are given as inputs to the Generalized Gamma Distribution proposed in setion II (A). The PDF of the is retrieved and stored. The query image is proessed using setion II (B). The relevant that are mathed are ompared against the PDF. The input query image is proessed using the methodology under-laid in Setion II (C) and the relevant are retrieved. The performane of the developed methodology is tested using metris like Preision, Reall, Error rate and Retrieval effiieny. In order to present this paper we have onsidered an image dataset obtained from University of Rennes. The relevant from the data set against the query are mathed based on Region of Interest (ROI) using the equation 5. The methodology is tested by varying the ranging ount from 00, 300, 500 and 000 in the database. The next level of analysis is to identify the math between the ROI and the onsidered with above sizes. Sine it is mandatory to test the appropriateness of the developed methodology, we have tested the performane on different operating systems by varying image sizes as mentioned above. The time elapsed against eah of the under different operating system environments like Windows, Linux, UNIX is onsidered. Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58

4 An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution 55 Personal Computer (PC) with Windows operating system with 00 in the database is onsidered and the result was suessful and the time elapsed is.8 seonds. The tests were onduted in the same environment on databases of sizes 300, 500 and 000 and the time elapsed for retrieval is.5 seonds,.75 seonds and.9 seonds respetively. The results were tested under the other operating systems with the above dataset and time taken is as follows. The table presents the effetiveness of the model where it is tested in different Operating System environments by varying the image sizes. In both the ases of operating Systems the model performs better retrieval auray in minimum time. Table. Performane of the system by varying the image sizes Operating System Linux UNIX Images Size Time elapsed in seonds The query medial image is onsidered and is to be mathed with the similar in the database, to identify whether the patient is normal or suffering from disease. A loal database is maintained and the query image onsidered is proessed with these in the loal database for relevane. The Query image is proessed based on Query based image tehnique presented in Setion II (B) and the relevant are obtained based on relevane method. These are presented in Figure and Figure 3. Disease Brain tumor Query image Relevant Non- Relevant Fig. 3. Input Query Image versus Retrieved Relevant and Non- Relevant A. Evaluation The outputs derived against the query proessed are evaluated using metris like Preision, Reall, Error rate and Retrieval effiieny. Preision It is the ratio of the number of relevant retrieved to the total number of irrelevant and relevant retrieved. It is usually expressed as a perentage. Preision = (A / (A + C))* 00; A: Number of relevant retrieved. C: Number of irrelevant retrieved. A + C: Total number of irrelevant + relevant retrieved. Reall It is the ratio of the number of relevant retrieved to the total number of relevant in the database. It is usually expressed as a perentage. Reall = (A / (A + B)) * 00 A: Number of relevant retrieved B: Number of relevant not retrieved A + B: The total number of relevant Error rate Error rate = Number of non-relevant retrieved / Total number of retrieved. Retrieval effiieny Retrieval effiieny = Preision = Number of relevant retrieved/ Total number of retrieved (If number of retrieved > number of relevant ), otherwise Retrieval effiieny = Number of relevant retrieved/ Total number of relevant. The outputs derived using the proposed metris are presented in Table and in Figure 4. Fig.. Image Dataset Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58

5 56 An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution Table. Preision, Reall, Error rate and Retrieval effiieny based on relevant and non-relevant No. of relevant in the database No. of nonrelevant in the database Ratio of relevane to nonrelevane Preision Reall Error rate Retrieval effiieny B. Analysis Of The Methodology Against The Time Taken Table 3 desribes the reognition auray against the time taken. From this table it an be learly seen that the relevant are retrieved effetively. Table 3. Showing the auray of Present model Brain tumor Brain Fratures Hemorrhage Auray (%) Time taken in seonds.8.0. C. Performane Evaluation The retrieved effiieny, in ase of medial against the query image is evaluated using Image quality metris. The various metris onsidered for the work inlude, Maximum distane, Mean Squared Error, Signal to Noise Ratio and Jaard quotient. The formulas for omputing the above quality metris are presented in Table 4. Using these formulae the performane evaluation is arried out and the proposed model is ompared to GMM and the results obtained are tabulated and presented in Table 5. Fig. 4. Variation of Preision, Reall, Error rate and Retrieval effiieny values with relevant to non-relevant ratio Table 4. Formulas for Quality Metris Quality metri Formula to Evaluate Maximum Distane Max{ F( k) F k) } ˆ( M N M N [ O{ F( k)} O{ Fˆ( k)}] j k j k MN Where M,N are image matrix rows and olumns [ O{ F( k)}] 0.log 0 MAX I MSE Where, MAX I is maximum possible pixel value of image, MSE is the Mean squared error Jaard quotient Where, X Y X Y a a b X Y a X Y, b,, d X Y Y X image intensities and X, Y are input and output Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58

6 An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution 57 Table 5. Evaluated Image Quality Metris Image Quality Metri GMM Maximum Distane Jaard quotient Maximum Distane Jaard quotient Generalized Gamma Distribution Standard Limits - to 0 to 0 to - to 0 to 0 to Standard Criteria Closer to Closer to Closer to Closer to Maximum Distane Jaard quotient to 0 to 0 to Closer to Closer to Maximum Distane Jaard quotient to 0 to 0 to Closer to Closer to IV. CONCLUSION In this paper an effiient methodology for brain MRI image retrieval against the query image is proposed. This proposed system helps to retrieve the relevant from the image dataset effetively. The proposed model exhibits good reognition rates of above 90%. The evaluation of the developed model is arried out by omparing with the existing models based on GMM, by using quality metris. The results show that, this developed algorithm outperforms the existing algorithm. This methodology suits very well in appliations wherein assistane is needed for the dotors at remote areas. REFERENCES [] Ahmad Fauzi M. F., and Ahmad W., Effiient blok based mathing for ontent based retrieval of CT head, In Proeedings of the IEEE 0th workshop on multimedia signal proessing, Cairns, Qld, pp. 4-47, 008. [] Babu G. P., and Murty M. N., A near-optimal initial seed value seletion in k-means means algorithm using a geneti algorithm, Pattern Reognition Letters, vol. 4, no. 0, pp , 993. [3] Bishnu P. S., and Bhattaherjee V., Software fault predition using quad tree-based -means lustering algorithm, IEEE Transations on nowledge and Data Engineering, vol. 4, no. 6, pp , 0. [4] Buiu I., and Gasadi A., Diretional features for automati tumor lassifiation of mammogram, Biomedial Signal Proessing and Control, vol. 6, no. 4, pp , 0. [5] Hamarneh G., and Li X., Watershed segmentation using prior shape and appearane knowledge, Image and Vision Computing, vol. 7, no., pp , 009. [6] Ibrahim S. I. Abuhaiba,Ruba A. A. Salamah,"Effiient Global and Region Content Based Image Retrieval", IJIGSP,vol.4,no.5,pp.38-46,0. [7] assner A., and Thornhill R. E., Texture analysis: a review of neurologi MR imaging appliations, Amerian Journal of Neuroradiology, vol. 3, no. 5, pp , 00. [8] Likas A., Vlassis N., and Verbeek J. J., The global k- means lustering algorithm, Pattern reognition, vol. 36, no., pp , 003. [9] Linde Y., Buzo A., and Gray R., An algorithm for vetor quantizer design, IEEE Transations on Communiations, vol. 8, no., pp , 980. [0] Li-xin S., Rui-feng C., and Qian W., Image retrieval of alifiation lusters in mammogram using feature fusion and relevane feedbak, In Proeedings of the international forum on strategi tehnology, pp 5-8, 00. [] Lu G., Tehniques and data strutures for effiient multimedia retrieval based on similarity, IEEE Transations on Multimedia, vol. 4, no. 3, pp , 00. [] Madhusudhana Rao T. V., Pallam Setty S., and Srinivas Y., Content based Image Retrievals for Brain Related Diseases, International Journal of Computer Appliations, vol. 85, no., pp , 04. [3] Mohamed M. Fouad,"Content-based Searh for Image Retrieval", IJIGSP, vol.5, no., pp.46-5, 03.DOI: 0.585/ijigsp [4] Muller H., Mihoux N., Bandon D., and Geissbuhler, A., A review of ontent-based image retrieval systems in medial appliations linial benefits and future diretions, International journal of medial informatis, vol. 73, no., pp. -3, 004. [5] MyungEun L., SoonYoung P., and WanHyun C., Medial image segmentation using geometri ative ontour model based on level set method, In Proeedings of the IEEE Paifi Rim Conferene on ommuniations, omputers and signal proessing, Vitoria, BC, pp , 007. [6] Ng H. P., Ong S. H., Foong. W. C., Goh P. S., and Nowinski W. L., Medial image segmentation using k- means lustering and improved watershed algorithm, In Proeedings of the IEEE Southwest symposium on image analysis and interpretation, Denver, CO, pp. 6-65, 006. [7] Prasad B. G., and rishna A. N., Statistial texture feature-based retrieval and performane evaluation of CT brain, In Proeedings of the 3 rd international onferene on eletronis omputer tehnology, anyakumari, India, pp , 0. [8] Quelle G., Lamard M., Cazuguel G., Roux C., and Cohener B., Case retrieval in medial databases by Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58

7 58 An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution fusing heterogeneous information, IEEE Transations on Medial Imaging, vol. 30, no., pp. 08-8, 0. [9] Rajkumar., and Muttan S., Medial image retrieval using energy effiient wavelet transform, In Proeedings of the 00 international onferene on omputing, ommuniation and network tehnologies (ICCCNT), arur, pp. -5, 00. [0] Sanjay S., and Trimbak S., An Effetive Mehanism to Neutralize the Semanti Gap in Content Based Image Retrieval (CBIR), The International Arab Journal of Information Tehnology, vol., no., pp. 4-33, 04 [] Shahabi C., and Safar M., An experimental study of alternative shape-based image retrieval tehniques, Multimedia Tools and Appliations, vol. 3, no., pp. 9-48, 007. [] Stoitsis J., Valavanis I., Mougiakakou S. G., Golemati S., Nikita A., and Nikita.S., Computer aided diagnosis based on medial image proessing and artifiial intelligene methods, Nulear Instruments and Methods in Physis Researh Setion A: Aelerators, Spetrometers, Detetors and Assoiated Equipment, vol. 569, no., pp , 006. [3] Trpovski Z., Content based image retrieval: from pixels to semantis, In Proeedings of the 9th symposium on neural network appliations in eletrial engineering, NEUREL, Belgrade, pp. 7-, 008. [4] World Health Organization Caner Fat Sheets, Available athttp:// ex.html, Aessed April 5, 04. Authors Profiles Mr. T.V. Madhusudhana Rao is urrently working as an Assoiate Professor in the Department of Computer siene and engineering at T.P.I.S.T., Bobbili. He reeived his B.Teh from JNT University, akinada, India and M.Teh from JNT University Anantapur, India. He is pursuing his PhD in the Department of Computer siene and engineering, at JNT University, akinada, India, in the area of Content based image retrievals. His other researh interests inlude Image Proessing, nowledge Disovery and Data Mining, Computer Vision and Image Analysis. He has published more than 0 papers in National/ International Conferenes and Journals. Dr. S. Pallam Setty is urrently working as a Professor in the Department of Computer Siene and Systems Engineering at Andhra University, Visakhapatnam, India. He reeived his PhD in Computer Siene and Systems Engineering from Andhra University, Visakhapatnam, India. He has 3 years of teahing and researh experiene. He has guided 4 students for Ph D and guiding sholars for Ph D. His urrent researh interests are in the areas of Image Proessing, omputer vision and image analysis, Computer Networks, Modeling and Simulation. He has more than 80 researh papers in various reputed international journals and onferene. Dr. Y. Srinivas is urrently working as a Professor in the Department of Information Tehnology at GITAM University, Visakhapatnam, India. He reeived his PhD in Computer Siene with Speialization in Image Proessing from Aharya Nagarjuna University, Guntur, India. He has 9 years of teahing and researh experiene. He has guided two students for Ph D and guiding eight sholars for Ph D. His urrent researh interests are in the areas of Image Proessing, nowledge disovery and data mining, omputer vision and image analysis. He has published more than 50 papers in various reputed international journals and onferene. How to ite this paper: T.V. Madhusudhana Rao, S.Pallam Setty, Y.Srinivas,"An Effiient System for Medial Image Retrieval using Generalized Gamma Distribution", IJIGSP, vol.7, no.6, pp.5-58, 05.DOI: 0.585/ijigsp Copyright 05 MECS I.J. Image, Graphis and Signal Proessing, 05, 6, 5-58