An Fuzzy Neural Approach for Medical Image Retrieval

Transcription

1 Journal of Computer Science 2012, 8 (11), ISSN doi: /jcssp Published Online 8 (11) 2012 ( An Fuzz Neural Approach for Medical Image Retrieval 1 C. Sriramakrishnan and 2 A. Shanmugam 1 Department of ECE, Adhiamaan College of Enginering, osur, India 2 Bannari Amman Institute of Technolog, Sath, Tamil Nadu, India Received , Revised ; Accepted ABSTRACT Image retrieval based on a quer image is necessar for effective and efficient use the information that is stored in medical image databases. Medical Image Retrieval is difficult as not onl the localization and directionalit of human visual sstem is to be considered but also the pathological condition. Image identification and segmentation for feature etraction pose a challenge to image retrieval process. Challenges posed include large number of images to be processed for the image retrieval and identifing the region of interest automaticall to optimize the search. In this stud, we propose a novel image segmentation algorithm Fuzz Edge Detection and Segmentation (FEDS). The proposed FEDS algorithm is tested on medical images and for classification of images, a bell fuzz multilaer perceptron is proposed. The proposed neural network Bell Fuzz Multi Laer Perceptron (BF-MLP) Neural network is constructed b introducing a fuzz logic in hidden laer with the sugeno model and bell function. The proposed neural network consists of two laers with the first laer being a tanh activation function and the second laer containing the bell fuzz activation function. The proposed FEDS method was implemented using Matlab and Modelsim. A total of 44 images were considered with three class labels. The edge obtained for which segmentation is done using the proposed segmentation algorithm. The proposed BF-MLP neural network algorithm was implemented using Visual Studio and the classification accurac compared with MLP Neural Network with sigmoid activation function. In this stud, a fuzz segmentation algorithm and a fuzz classification algorithm is proposed to improve the medical image retrieval accurac. The proposed segmentation algorithm, Fuzz Edge Detection and Segmentation (FEDS), was implemented using Matlab and features were etracted using Fast artle Transform (FT). The etracted features were used to train the proposed neural network, Bell Fuzz Multi Laer Perceptron Neural Network (BF-MLP). 44 images with 3 class labels were used to test the algorithm and classification accurac of 93.2% was obtained. Kewords: Content Based Image Retrieval (CBIR), Medical Image, Fuzz Edge Detection and Segmentation (FEDS), Bell Fuzz Multilaer Perceptron 1. INTRODUCTION With the availabilit of large amount of digital medical images in databases, image retrieval based on a quer image is necessar for effective and efficient use of the information available. Content Based Image Retrieval (CBIR) is the process of locating similar picture/pictures in an image database when a quer image is provided. When the picture of a car is given, the sstem should retrieve similar car images in the database. Automated etraction of features from the image is crucial for image retrieval sstems (Rui et al., 1999). This is achieved through etracting image features like color, teture and shape. Such image features are then compared to the quer image and database images. A similarit algorithm calculates the similarit degree between both images. Database images which have same features as the quer image (with the highest similarit measure) are then ranked and presented to the user. CBIR is a two step process which includes Feature Etraction and Image Matching (also called feature Corresponding Author: Sriramakrishnan, C., Department of ECE, Adhiamaan College of Enginering, osur, India Tel:

2 matching). Feature Etraction etracts image features to a discerning etent and information like color, teture and shape etracted from images are called feature vectors. Etraction is undertaken on quer and database images. Image matching includes use of features of both images and comparison between them to ensure features similar to those in database images. Image segmentation represents a digital image in simple multiple region forms to facilitate analsis. Objects, boundaries in an image are sited and also object and background is segmented using image segmentation (Gonzale and Woods, 2009; Banerjee and Kundu, 2003; Amartur et al., 1992; Lo et al., 1995; ussain et al., 2012; Barskar and Ahmed, 2011). Generall, image segmentation process first segments color which is followed b edge detection. Feature etraction depends upon the qualit of segmentation as over segmentation leads to man details of the object and under segmentation groups man object into single region. So level of segmentation required is a deciding factor for success or failure of analsis. The image segmentation algorithm takes advantage of discontinuit and similarit of the image gre levels. Some of the most commonl used segmentation techniques are histogram thresholding, clustering, vector based and fuzz technique. In this stud, a novel image segmentation algorithm Fuzz Edge Detection and Segmentation (FEDS) is proposed. The proposed FEDS algorithm is tested on medical images and for classification of images, a bell fuzz multilaer perceptron is proposed. The rest of the stud is organized as follows: Section 2 reviews fuzz techniques used in CBIR available in literature, section 3 details about the proposed methods, section 4 gives the eperimental details and results and section 5 concludes the stud. 2. MATERIALS AND METODS The dataset consists of 44 medical images from the National Biomedical Imaging archive for the eperimentation of image retrieval. Images were segmented using the proposed Fuzz Edge Detection Segmentation (FEDS) algorithm. Features were etracted from the segmented images using Fast artle Transform. A bell fuzz multilaer perceptron is proposed for image classification Fuzz Edge Detection and Segmentation (FEDS) Edge detection is a process of identifing and locating edges in image for image segmentation. Distortion, noise, overlaps, intensit variation are some of the factors which contributes to edge etraction (Carron and Lambert, 1995). There are man methods in literature for edge detection; fuzz method is used in this stud. Fuzz theor is effective for modeling ambiguit or uncertaint, thus, the noise is efficientl removed from the image when detecting edges. The image is transformed into fuzz image b the use of fuzzification function (Re-Sern, 2008). In fuzz segmentation, the regions in the image are obtained b connecting and merging the constant color inside image. Propinquit, a metric for closeness between components, defined b the fuzz inference sstem is used. The propinquit measure depends on the color distance and the topological relation between components. The fuzz segmentation method constructs clearer segmentation. In the proposed algorithm, fuzz logic is used to find the edge based on the output of the two edge detection method. Initiall the edge is detected using Sobel operator. The Sobel operator filter kernel is used and is given b Eq. 1 and 2: (1) (2) The vertical and horizontal gradient component and are obtained b the above kernels. The gradient magnitude is approimated as follows Eq. 3: + (3) And the angle of the orientation of the edge relative to the piel grid is given b Eq. 4: ϕ a rc ta n (4) The two kernels can be simultaneousl applied on image to obtain the approimate magnitude as follows Eq. 5 and 6: a1 a2 a3 a 4 a5 a6 a 7 a8 a 9 (5) 1810

3 ( a1+ 2a 2+ a3) ( a7+ 2a8+ a9) ( a 2a a ) ( a 2a a ) (6) In the second method the Laplacian operator kernel given b Eq. 7 is used: I The magnitude of the image is given b Eq. 8: ( 5 ( ) ) (7) I 4a a + a + a + a (8) The magnitudes from both the methods are normalized to obtain values in the range [0, 1]. The normalized outputs are mapped to three class values V l, Vm and V h from Fig. 2 the range of the class values are given b: t V 0,0.8 m V 0, 0.8 h V 0.8,1 The output of the fuzz sstem determines whether the piel is an edge or not based on the defined fuzz rules. The three classes of outputs are E l, E m and E h which defines the probabilit of the piel being an edge is low, medium or high. If p 1 and p 2 are the corresponding values from the two edge detection methods, some of the fuzz rules can be described b: If p 1 in V 1 and p 2 in V k then the probabilit of edge is classified to E m If p 1 in V h and p 2 in V h then the probabilit of edge is classified to E h A total of 9 combinations are possible with generation of 9 rules. The outer edge value is used to create a mask and features are etracted using Fast artle Transform. The artle transforms maps real-valued temporal or spacial functions into real-valued frequenc function decomposing into two independent sets of sinusoidal components (artle, 1942). Energ features are etracted using the Fast artle Transform on the etracted segments. The obtained energies are used to train the neural network. Fig. 1. Block diagram showing preprocessing fuzz logic to neural network Table 1. Design metrics of the proposed neural network model Input Neuron 20 Output Neuron 1 Number of idden Laer 2 Number of processing elements-first laer 6 Transfer function of first hidden laer tanh Learning rule momentum Number of processing elements-second laer 2 Transfer function of second hidden laer bellfuzz Learning Rule of hidden laer Momentum 2.2. Multilaer Perceptron Neural Network Multilaer Perceptrons (MLPs) are feed forward neural networks; these are supervised networks and widel used for modeling prediction problems (ornik et al., 1989). During training the network learns how to transform input data into a desired response, the MLPs are generall trained with the standard back propagation algorithm. Back propagation computes the sensitivit of the output with respect to each weight in the network and modifies each twilight b a value that is proportional to the sensitivit. In training process, the representative eamples are iterativel presented to the network, so that it can integrate this knowledge within its structure. A good training algorithm not onl shortens the training time but also achieves better accurac. There are a number of was fuzz logic can be used with neural networks. The fuzz logic is used in the form of a fuzzifier function to preprocess or post-process data for a neural network as shown in Fig. 1. In the second method fuzziness can enter neural networks to define the weights from fuzz sets Proposed Neural Network Model 3.2. Bell Fuzz Multi Laer Perceptron Neural Network (BF-MLP Neural Network) The proposed neural network Bell Fuzz Multi Laer Perceptron (BF-MLP) Neural network is constructed b introducing a fuzz logic in hidden laer with the sugeno model and bell function. The proposed neural network 1811

4 consists of two laers with the first laer being a tanh activation function and the second laer containing the bell fuzz activation function. The Bell Fuzz Multi Laer Perceptron (BF-MLP) Neural network proposed in this stud uses the criteria specified in Table 1. The Sugeno fuzz model is the fuzz model introduced in the proposed algorithm. A tpical fuzz rule in Sugeno fuzz model has the form If is A and is B then z f (, ).where A and B are fuzz sets in the antecedent part, while z f (, ) is crisp function in the consequent part (Takagi and Sugeno, 1985). The fuzz bell membership function is given b: The proposed BF-MLP neural network algorithm was implemented using Visual Studio and the classification accurac compared with MLP Neural Network with sigmoid activation function. Results of the classification accurac obtained are displaed in Fig. 4 and 2. Table 2 and 3 tabulates the precision, recall and f Measure of the various images used in the eperiment. Figure 5 and 6 shows the precision, recall and f Measure of the various images used in the eperiment w w w i where, is the input and w i is the weight. 3. RESULTS AND DISCUSSION The proposed FEDS method was implemented using Matlab and Modelsim. A total of 44 images were considered with three class labels. Images used in the eperimental setup are shown in Fig. 2 and 3 shows the edge obtained for which segmentation is done using the proposed segmentation algorithm. Fig. 4. Classification accurac obtained Fig. 2. Images used in the eperiment Fig. 5. Precision and Recall Fig. 3. Input image with obtained edges from proposed method Table 2. Classification Accurac Classification Technique used accurac % FEDS-MLP 90.9 FEDS-proposed bell fuzz MLP

5 Fig. 6. f Measure Table 3. Summar of proposed fuzz classifier and FEDS Classifi- FEDS-MLP FEDS-proposed bell fuzz MLP Cation Accurac 90.90% 93.20% Class Precision Recall Fmeasure Precision Recall Fmeasure chest_ image BrainA BrainB CONCLUSION In this stud, a fuzz segmentation algorithm and a fuzz classification algorithm is proposed to improve the medical image retrieval accurac. The proposed segmentation algorithm, Fuzz Edge Detection and Segmentation (FEDS), was implemented using Matlab and features were etracted using Fast artle Transform (FT). The etracted features were used to train the proposed neural network, Bell Fuzz Multi Laer Perceptron Neural Network (BF-MLP). 44 images with 3 class labels were used to test the algorithm and classification accurac of 93.2% was obtained. 5. ACKNOWLEDGEMENT The authors are etremel thankful to Computational Visual Cognition Laborator and websites mentioned in the references. 6. REFERENCES Amartur, S.C., D. Piraino and Y. Takefuji, Optimization neural networks for the segmentation of magnetic resonance images. IEEE Trans. Med. Imag., 11: PMID: Banerjee, M. and M.K. Kundu, Content based image retrieval with fuzz geometrical features. Proceedings of the 12th IEEE International Conference on Fuzz Sstems, Ma 25-28, IEEE Xplore Press, pp: DOI: /FUZZ Barskar, R. and G.F. Ahmed, CBIR using fuzz edge detection mask. Proceeding of the 2011 International Conference on Communication, Computing and Securit, (ICCCS 11), ACM New York, USA., pp: DOI: / Carron, T. and P. Lambert, Fuzz color edge etraction b inference rules quantitative stud and evaluation of performances. Proceedings of the International Conference on Image Processing, (ICIP 95), ACM Press, USA., pp Gonzale, R.C. and R.E. Woods, Digital Image Processing. 3rd Edn., Pearson Education India, India, ISBN-10: , pp: 954. artle, R.V.L., A more smmetrical fourier analsis applied to transmission problems. Proc. IRE, 30: DOI: /JRPROC ornik, K., M. Stinchcombe and. White, Multilaer feedforward networks are universal approimators. Neural Netw., 2: DOI: / (89) ussain, S.J., T.S. Savithri and P.V.S. Devi, Segmentation of tissues in brain MRI images using dnamic neuro-fuzz technique. Int. J. Soft Comput. Eng., 1: Lo, S.C.B., S.L.A. Lou, J.S. Lin, M.T. Freedman and M.V. Chien et al., Artificial convolution neural network techniques and applications for lung nodule detection. IEEE Trans. Med. Imag., 14: DOI: / Re-Sern, L., Edge detection b morphological operations and fuzz reasoning. Proceedings of the 2008 Congress on Image and Signal Processing, (CISP 08), ACM Press, USA., pp: DOI: /CISP Rui, Y., T.S. uang and S.F. Chang, Image retrieval: Current techniques, promising directions and open issues. J. Vis. Commun. Image Represent., 10: Takagi, T. and M. Sugeno, Fuzz identification of sstems and its applications to modeling and control. IEEE Trans. Sst. Man Cbern., 15: