MEDICAL IMAGE RETRIEVAL BY COMBINING LOW LEVEL FEATURES AND DICOM FEATURES

Size: px

Start display at page:

Download "MEDICAL IMAGE RETRIEVAL BY COMBINING LOW LEVEL FEATURES AND DICOM FEATURES"

Vivian Carpenter
5 years ago
Views:

1 International Conference on Computational Intelligence and Multimedia Applications 2007 MEDICAL IMAGE RETRIEVAL BY COMBINING LOW LEVEL FEATURES AND DICOM FEATURES A. Grace Selvarani a and Dr. S. Annadurai b a. Senior Lecturer, Sri Ramakrishna Engineering College, Coimbatore, India. Id : dg_jane@yahoo.co.in b. Principal, Government College of Engineering, Tirunelveli, India. Id : anna_prof@yahoo.co.in Abstract Content based image retrieval aims at searching the image databases using non-textual information which are low level features like color, shape and texture. Medical image databases contain lot of textual or semantic information. This paper presents an approach by combining low level content features and high level semantic features to perform retrieval on medical image databases. The semantic information is extracted from DICOM header which is used to perform the initial search and images are retrieved. This pre-filtering of the images reduces the number of images to be searched. Content-based retrieval is performed only to the pre filtered image database which speeds up the retrieval process. Retrieval is performed by extracting shape and texture features. Experimental result shows that by combining the high level semantics (DICOM features) and low level content features (shape and texture) the retrieval time is reduced and the performance of medical image retrieval is increased. Keywords: CBIR, DICOM, Gabor wavelet, medical images. 1. Introduction A large number of medical images in digital format are generated by hospitals and clinics every day. Such images constitute an important source of anatomical and functional information for diagnosis of diseases, medical research and education. It is well acknowledged that medical image databases are a key component in diagnosis and preventive medicine. This increasing trend towards digitization of medical images creates a need of technologies for storage, organization and retrieval of the medical images. Content based image retrieval (CBIR) is the digital image searching problem in large databases that makes use of the contents of the images themselves rather than relying on the textual information. These techniques use the automatically derived features such as color, texture and shape as search criteria. Medical images generated in hospitals contain semantic information. This information can be used to retrieve the images. The unique characteristics of medical images hinder the direct adaptation of contentbased retrieval approaches, which are already in use for unstructured collections of images. Early systems, such as Photobook [4] or the Query By Image Content (QBIC) of IBM Inc. [5] model only a rudimentary understanding of image content. Most systems for medical image retrieval are text-based [6,7,8], strongly rely on manual input [9], and/or need to be tailored for a specific application and modality [10] /07 $ IEEE DOI /ICCIMA

2 The deep gap between low-level features and high-level semantics concepts is a major obstacle to more effective image retrieval. Existing systems still fail when applied to medical image databases because of the rich semantics. The main objective of this work is to combine the low level features and high level semantics. 2. Proposed System The proposed system uses the DICOM information for performing the retrieval on medical images. To reduce the gap between the low level features and semantic features, retrieval is performed by combining low level features (shape and texture) and semantic information extracted from the dataset values of the DICOM format. The extracted information can be used to perform the initial retrieval which produces a set of images which are relevant to the given query image. This pre-filtering of the image databases reduces the number of images to be searched. From the set of images obtained after searching through the dataset values, retrieval is performed using low level features. In the proposed system we use shape and texture features. The fixed resolution block representation shape extraction algorithm is used. This application makes use of the Gabor wavelet for texture extraction. Similarity between the query image features and the features of the images in the database has been calculated. The N most similar images are retrieved based on the similarity criteria. 3. Semantic Feature Extraction The Digital Imaging and Communications in Medicine (DICOM) standard was created by the National Electrical Manufacturers Association (NEMA) to aid the distribution and viewing of medical images, such as CT scans, MRIs, and ultrasound. Imaging equipment used in hospitals generates images which are in DICOM format. It is a standard format used to obtain, store and distribute medical images. DICOM comprise standardized textual descriptions of study, patient, body region examined and modality. A single DICOM file contains both a header (which stores information about the patient's name, the type of scan, image dimensions, etc), as well as all of the image data. This is different from the popular Analyze format, which stores the image data in one file (*.img) and the header data in another file (*.hdr). The DICOM header size varies depending on how much header information is stored. The header describes the image dimensions and retains other text information about the scan. DICOM files are composed by one image and tags describing the image. Tags are textual or numerical sequences of <attribute, value> pairs. The textual information are considered as the semantic information. For all the DICOM files the image and the relevant tags are extracted and are stored in the database. The image is stored in jpeg file format. The extracted semantic information is stored in the database which is used during the retrieval process. 4. Content Feature Extraction Content Based Retrieval system represent each image as a feature vector and measures the similarity between images as the distance between their corresponding feature vectors. For medical images shape and texture are the two important low level features which describe the content of the image. The shape and texture features are extracted and stored in the database as feature vectors. 4.1 Shape Feature Extraction 588

3 The fixed block resolution format [1] is used for shape representation and extraction of shape features. The details of the extraction step are described as below. Step 1 Divide the image into isometric blocks that contain N x N pixels. Step 2 Judge whether there are over p % (p is between 1~100) pixels greater than a certain critical value in each block; if it is true, the index of this block will be set to 1, if it is not, it will be set to 0. Step 3 Judge whether the shapes of the two objects are similar; comparing the block index produced in step 2, if they are different, then add 1 to the result; the smaller result represents that the shapes are more similar. For example, an image with 256 x 256 pixels, if N equals to 32, the image is divided into 8 x 8 blocks, each block contains 32 x 32 pixels; if N equals to 16, the image is divided into 16 x 16 blocks, each block contains 16 x 16 pixels. Therefore, if there are more blocks, in spite of more data quantity, more specific object shape can be recorded and the resolution is higher. 4.2 Texture Feature Extraction Texture is the key component of human visual perception. Everyone can recognize texture, but it is more difficult to define. Texture occurs over a region rather that at a point. Gabor filter [2] is one of the most popular signal processing based approach for texture extraction. Basically, Gabor filters are a group of wavelets, with each wavelet capturing energy at a specific frequency and specific direction. An image is filtered with a set of Gabor filters with different preferred orientations and spatial frequencies and the features, which are obtained from a feature vector, is used further. Texture features are found by calculating the mean and variation of the Gabor filtered image.for a given image I(x,y) with size PxQ, its discrete Gabor wavelet transform is given by a convolution: (1) where s and t are filter mask size variables and Ψ * mn is the complex conjugate of Ψ mn which is a class of self-similar functions generated from dilation and rotation of the following mother wavelet: (2) After applying Gabor filters on the image with different orientation and at different scale, an array of magnitudes are obtained: (3) These magnitudes represent the energy content at different scale and orientation of the image. Texture feature are found by calculating the mean µ mn and standard deviation σ mn of the energy magnitude 589

4 (4) A feature vector f which represents the texture is created using µ mn and σ mn as the feature components. In the implementation we have used 4 scales and 6 orientations and the feature vector is given by f = ( µ 00, σ 00, µ 01, σ 01, µ 35, σ 35 ) (5) 5. Retrieval Using Dataset And Content Features Retrieval is the process of searching for images that matches the given input image. Before the retrieval, the semantic information, texture and shape features are extracted for all images in the database. When the query image is given its semantic features are extracted and are compared with the semantic features of all the images in the database. In the proposed system the textual information related to body region examined and modality are used for performing the match. From the set of images retrieved, content based search is performed. Content based search is performed by computing the similarity measurement which takes into account the shape and texture features. Euclidean distance between the shape feature of the query image and the each image in the databases is computed. Similarly Euclidean distance between the texture feature of the query image and each image in the database is calculated. The similarity function S(q,i) is given by S(q,i) = D S (q,i) + D T (q,i) (6) Where q is the query image and i is any image in the database. D S (q,i) is shape distance between the query image and image in the database. D T (q,i) is texture distance between the query image and image in the database. S(q,i) gives the similarity value between the query image and image in the database. Rank the images according to the decreasing order of similarity value. The top N similar images are retrieved according to the rank. 6. Experimental Results Experiments were conducted to show the performance of the system. In the implemented system first, the semantic information is extracted and retrieval is performed which speeds the retrieval process. Experiments were conducted with and without combining the semantic information. Results show that retrieval time is considerably reduced when semantic information is used for retrieval. To evaluate the performance of the proposed scheme for image retrieval, the commonly used factor, Precision is used to represent the retrieval accuracy which is the percentage of similar images retrieved with respect to the number of retrieved images. Precision is calculated as the average of 20 queries. Results show that the retrieval accuracy has been increased when semantic information is combined with content features. Table 6.1 shows the precision values. Table 6.1 Average precision values No of retrieved images Precision

5 This system have been implemented using JDK1.5 in a PIV processor ( 512MB RAM, 3GHz ). The database consists of 10,000 medical images of different modalities and various parts of the body. 7. Conclusion In this paper a medical image retrieval system which combines the semantic information and content features have been proposed to reduce the gap between the low level features and semantic features. Retrieval is performed first by extracting semantic information from the dataset values of the DICOM format which produces a set of images which are relevant to the given query image. This pre-filtering of the image databases reduces the number of images to be searched. From the set of images obtained after searching through the dataset values, retrieval is performed using low level features (shape and texture). Thus, it reduces the time taken to search the entire medical image database. Also the accuracy of retrieval is improved by this method. The future work is to combine the user s preference using the relevance feedback technique. 8. References [1] Jann-Der Lee and Li-Peng Lou, Using Texture and Shape Features To Retrieve Sets of Similar Medical Images, Vol 15, No 5, pp , Oct [2] A.Grace Selvarani and Dr. S. Annadurai, Medical Image Retrieval using Gabor Wavelet, i-manager s Journal on software Engineering, Sept [3] Kwok-Wah Hung and Aw-Yong M, "A Content based Image Retrieval System Integrating Color, Shape and Spatial Analysis", IEEE International Conference on Systems Man and Cybernetics, vol.2, pp , [4] A. W. M. Smeulders, M. Worring, S. Santini, A. Gupta, and R. Jain. Content-based image retrieval at the end of the early years IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(12):1349{1380, [5] Flicker M, Sawhmey H, Niblack W, Ashly J, et al.: Query be image and video content the QBIC system. IEEE Computer 28(9):23 32, [6] Long LR, Berman LE, Thoma GR: A prototype client/server application for biomedical text/image retrieval on the Internet. Procs. SPIE 2670:362 72, [7] Long LR, Pillemer SR, Lawrence RC, Goh GH, Neve L, Thoma GR: WebMIRS Web-based medi-cal information retrieval system. Procs. SPIE 3312: , [8] Lowe HJ, Antipov I, Hersh W, Smith CA, Mailhot M: Automated semantic indexing of imaging reports to support retrieval of medical images in the multimedia electronic medical record. Methods of Information in Medicine 38: , [9] Liu Y, Rothfus WE, Kanade T: Content-based 3D neuroradiologic image retrieval preliminary results. Technical Report CMU-RI-TR-98-04, Carnegie Mellon University, Pittsgurgh, PA, [10] Shyu CR, Brodley CE, Kak AC, Kosaka A, Aisen AM, Broderick LS: ASSERT a physician-in the loop contentbased retrieval system for HRCT image databases. Computer Vision & Image Under-standing 75: ,

CHAPTER 6 PROPOSED HYBRID MEDICAL IMAGE RETRIEVAL SYSTEM USING SEMANTIC AND VISUAL FEATURES

188 CHAPTER 6 PROPOSED HYBRID MEDICAL IMAGE RETRIEVAL SYSTEM USING SEMANTIC AND VISUAL FEATURES 6.1 INTRODUCTION Image representation schemes designed for image retrieval systems are categorized into two