Content Based Image Retrival of MRI of
Human Brain
A. Dr. Qaim Mehdi Rizvi, B. Dr. A. A. Karawia Department of Computer Science, Deanship of Education Services
sirqaim@gmail.com
Qassim University, Al-Qassim, Kingdom of Saudi Arabia. abibka@mans.edu.eg
Abstract
Medical imaging has become one of the most important clinical diagnosis components. Medical images are different from general images in the sense that they contain anatomical expertise. However, general registration techniques do not well exploit such expertise so better extraction and matching techniques are required for precise computer assisted diagnosis. Content-based image retrieval (CBIR) of medical images is an important alternate and complement to traditional text-based retrieval using keywords. The purpose of this Paper is to describe our research on improvement of feature extraction and matching technique for more precise and accurate Content Based Image Retrieval (CBIR) system specially designed for Brain scan images regarding various brain diseases and abnormalities.
Keywords: Computer Assisted Diagnosis, Mean Weight Factor, Computed Tomography, Cut-off Limit, Threshold, Entropy.
I. INTRODUCTION
With the rapid development of technology, the traditional information retrieval techniques based on keywords are not sufficient purposefully Content-based image retrieval has been a vigorous area of research for at least the last two Decades. Content Based Image Retrieval (CBIR) technologies provide a method to find images in large databases by using unique descriptors from a trained image. It plays a central role in the application areas such as multimedia database systems in recent years. The work focused on using low-level features like color, texture, shape and spatial layout for image representation (Fuhuri et al,2003).
II. PROPOSED WORK
In CBIR each image that is stored in the database has its features extracted and compared to the features of the query image. It involves two steps:
• Feature Extraction
• Texture Matching
The above steps described in detail in our next section III and section IV. The CBIR system will work upon two types of images:
• Query Image or Input Image
• Database (DB) Images: These images will be preprocessed before performing comparison among DB Images and the Query Image.
Preprocessing of both types of images consists of following steps:
1. Converting all images into gray scale images 2. Resizing all gray scale images into 400x400 scale 3. Quantizing all images into ten gray levels (figure-1) 4. Mean frequency for each quantized image computed
Figure.1. Quantization of image a2
Here the DB images is of various Brain diseases CT scan images
III. FEATURE EXTRACTION
The first step in this process is to extract the image features to a distinguishable extent. The feature is defined as a function of one or more measurements, each of which specifies some quantifiable property of an object, and is computed such that it quantifies some significant characteristics of the object (zhang et al,206). We classify the various features currently employed as follows:
General features: Application independent features such as color, texture, and shape. According to the abstraction level, they can be further divided into:
Pixel-level features: Features calculated at each pixel, e.g. color, location.
Global features: Features calculated over the entire image or just regular sub-area of an image.
Domain-Specific Features: Application dependent features such as human faces, fingerprints, and conceptual features.
These features are often a synthesis of low-level features for a specific domain.
On the other hand, all features can be coarsely classified into low-level features and high-level features. Low-level features can be extracted directed from the original images, whereas high-level feature extraction must be based on low-level features.
IV. MATCHING
The second step involves matching these features to yield a result that is visually similar. As result of Feature extraction, a vector of each image sample on the database is produced (Birgale et al,2006). The produced vectors then will be saved in the memory in indexing feature vector, which contains the indexes to both the names and the images of training database (Zhang et al,2004). The same process is followed for the query image and a vector of same dimensions as for the training database is produced. This is because we are using same reduction of dimensionality technique (PCA) for the query image as well. Cluster analysis and Outlier analysis is also very important to figure out the similarity between features of images at the time of classification. Steps to perform Matching can be given as below:
1. Level by level comparison between database image and input image is done only for those frequencies which are above the computed mean frequency
2. Matching weight factor increases as the difference between the frequencies of both the images falls within the range of cut off threshold
3. Matching weight factor sorted and corresponding images is shown in result. Here in figure-2, a2 is query image and comparison is being done with DB image a8. For Texture matching histogram in figure-3 and figure-4 is plotted for each of the two. Its x-axis shows ten gray levels and y-axis shows frequency per bin (gray level).
Figure.3. histogram for quantized image a8
Figure.4. histogram for quantized image a2
Similarly Mean weight factor for each of the DB image will be calculated and stored. Finally all MWFs will be sorted and DB images of MWFs greater than predefined threshold will be displayed in results.Deciding cut off limit is quite critical and results keep varying as it changes.
Flow chart:
The flowchart for the whole CBIR process for Brain CT scan images is shown in Figure-6
V. RESULTS AND SYSTEM PERFORMANCE EVALUATION
Testing the correctness and effectiveness of the system evolved becomes more important when the system is going to be used in real- time and more when it is for mankind i.e. medical diagnosis. Determining whether or not two images are similar is purely up to the user's perception. Human perceptions can easily recognize the similarity between two images although in some cases, different users can give different opinions but when computer assisted diagnosis is considered then more care is needed before recommending any system. Different evaluation measures were used here to evaluate the effectiveness of the image search engine system. These are as under:
i. Precision
It is a measure of the ability of a system to present only relevant items Precision=[total relevant images retrieved/total images retrieved]
ii. False Acceptance Ratio
For each query, the system returns the x closest images to the query, including the query image itself. The number of mismatches can be computed as the number of images returned that belong to a class different than that of the query image.
False Acceptance Ratio=[mismatch/x]
iii. False Rejection Ratio
The numbers of images that belong to the query image class, but that have not been returned by the system i.e. mismatch
False Rejection Ratio=[mismatch/x]
iv. Retrieval Score
The retrieval score for one class or query image can be then computed as Retrieval Score=[1-( mismatcha + mismatchr)/x] x 100 %
Figure.8. Sample DB images
Figure.9. Search results
For sample database of 8 images what result the system will give, the effectiveness and accuracy can be measured with the help of search results (figure-9) as under:
Recall=4/4 Precision=4/5
False acceptance ratio=1/5 False rejection ratio=0 Retrieval score=80%
Here, threshold=2 MWF (predefined) Cut off limit=± 2500 (predefined)
VI.FUTURE WORKS
VII. CONCLUSION
In this paper, the outcome of feature extraction and matching by setting cut off limit and threshold is pretty promising. Further studies can be done apart from CT scan for more generalized CBIR system. We show that our method is able to search for Brain CT scan images more accurately and efficiently. The proposed method can be used as a component in real-time CAD or CBIR system (Zhang et al,2009).
APPENDIX Matlab Code i=imread('d:\projfolder\alzg\alz2g.jpg'); %imshow(i); x=i; c=i; d=rgb2gray(x); imshow(d); x=d; for m=1:400 for n=1:400 if(x(m,n)<15) x(m,n)=0;
elseif(x(m,n)>=15 && x(m,n)<40) x(m,n)=25;
elseif(x(m,n)>=40 && x(m,n)<65) x(m,n)=50;
elseif(x(m,n)>=65 && x(m,n)<90) x(m,n)=75;
elseif(x(m,n)>=90 && x(m,n)<=115) x(m,n)=100;
elseif(x(m,n)>=115 && x(m,n)<140) x(m,n)=125;
elseif(x(m,n)>=140 && x(m,n)<165) x(m,n)=150;
elseif(x(m,n)>=165 && x(m,n)<190) x(m,n)=175;
elseif(x(m,n)>=190 && x(m,n)<215) x(m,n)=200;
References
i. Dr.Fuhuri Long,Dr.Hongjiang and Prof .David Dagan Feng, 2003.Fundamental of
contentbased ImageRetrieval
ii. Smeulders, A. W. M., Worring, M., Santini, S., Gupta, A.,and Jain,V.,2000. Content-Based Image Retrieval at the End ofthe Years, IEEE Transactions on Pattern Analysis and Machine Intelligence 22, pp. 1349–1380
iii. S.Li, T.Gong, J.Wang, R.Liu, C.L.Tan, T.Y.Leong, B.C.Pang,C.C.T.Lim,
C.K.Lee, Q.Tian, and Z.Zhang, “TBIdoc:3D Content-based CT Image Retrieval System for Traumatic Brain Injury”, To appear in SPIE Medical Imaging, San Diego, CA,US, 2010.
iv. zhen-hua zhang, yong quan, wenhui li, wu guo, “a new contentbased image retrieval”, Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006.
v. Lenina Birgale, Manesh Kokare and Dharmpal Doye, “Colour and Texture Features for Content Based Image Retrieval”, Proceedings of the International Conference on Computer Graphics, Imaging and Visualisation (CGIV'06) 0-7695-2606-3/06 $20.00 © 2006IEEE.
vi. Dengsheng Zhang and Guojun Lu ,“evaluation of similarity Measurement for image Retrieval”, 1-4240-2407-8/08/$25.00 ©2004 ieee
vii. Bin Zheng, “Computer-Aided Diagnosis in Mammography Using Content-Based