Aura Tranformcbir
Aura Tranformcbir
Aura Tranformcbir
1, February 2012
Abstract
The quality of biomedical images can be enhanced by using several transformations reported in the
literature. The enhanced images may be useful to extract the exact information from these scans. In recent
times, researchers exploited various mathematical models to smoothen and enhance the quality of the
biomedical images with an objective to extract maximum useful medical information related to functioning
or malfunctioning of the brain. Both real and non-real time based techniques have been explored and
reported for this purpose. In this proposed work, aura based technique has been investigated for enhancing
the quality of magnetic resonance imaging (MRI) scans of the human brain. The aura transformation based
algorithm with some modifications has been developed and the performance of the algorithm is evaluated
on a series of defected, diseased, and normal MRI brain images.
Keywords : MRI, image segmentation, brain tumor, aura transition, image processing.
I. INTRODUCTION
Image processing is a significant technique in several areas such as video coding, computer
vision, and medical imaging [1] [2]. In medical imaging, for the successful treatment of lesions,
tumors and vascular malformations in brain, MRI scans are vital. For further investigations, the
deformable model images and projection of structures from brain onto MRI image is a commonly
used [3]. These scans help in the diagnosis of injury for improving the results of surgical, radiosurgical, or radio-therapeutic treatment [4]. Converting an image into homogeneous regions is an
area of research from a decade, especially when the image is made up of complex textures. A
number of techniques have been proposed for this task, including spatial frequency techniques [5]
[6]. Image processing techniques have been used widely depending on the specific application
and image modalities. Computer based detection of abnormal growth of tissues in a human brain
are preferred to manual processing methods in the medical investigations because of accuracy and
satisfactory results. In MRI scans, the important task is to label the voxels according to their
tissue type which includes white matter (WM), gray matter (GM), cerebrospinal fluid (CSF), and
sometimes pathological tissues like tumor [7] [8]. Several methods for processing the MRI scans
DOI : 10.5121/sipij.2012.3109
95
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
have been developed. The different methods of analyzing the scans can be classified under five
broad categories. These are methods based on statistics (clustering methods) [9], fuzzy sets theory
[10], mathematical morphology [11], edge detection [12], and region growing [13]. Image
processing of a brain image allows extracting the lesions from the brain data and provides
information for further stages of the quantitative assessment. A lot of techniques have been
proposed for processing MRI scans for the purpose of effective diagnosis [14-16]. However, there
is still scope for improvement in terms of quality of the processed images. As a lot of information
is embedded in the texture of the MRI scans, the analysis of the texture of MRI scan may be quite
significant for predicting the brain related problems.
Recently, a technique called aura introduced by Rosalind W. Picard and Ibrahim Elfadel [17-19]
has been reported and applied for the analysis of the textures, restoration of the distorted images
and segmentation of geometrical patterns [20-25]. Even though many techniques have been used
for analysis of MRI images, the extensive use of aura based techniques for enhancing the content
of the scans has not been carried out.
In this research work, an aura based technique is investigated for enhancing the quality of an MRI
scans for better medical diagnosis. Extensive investigations have been carried out with the MRI
scans of several subjects involving different types of brain related problems. The analysis of the
processed images indicates the enhancement of the important regions of the brain in the MRI scan
responsible for the abnormalities in the functioning of the brain. For employing the standard
model reported by Rosalind W. Picard and Ibrahim Elfadel [17-19], to cover gray scale images
efficiently, few modifications have also been incorporated. The details of MRI have been
presented in the next section. In the subsequent sections, Aura transformation, mathematical
background of aura, methodology, and results are discussed.
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
useful for the diagnosis and treatment in brain pathology. MRI techniques are still in the
development stage, and most of the research is concentrated around the improvement of the
quality and speed of acquisition of MRI signal. Several processes involved in the interpretation of
MRI signals are manual. Therefore, even in the laboratories having the state of art technologies
there is a necessity for automated image processing technique for MRI based medical diagnosis.
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
image. After placing this element at a particular site, the intensity values taken from two sets
called Set A and Set B are matched with the pattern of the pixel values of the image just below
the structuring element. The choice of the shape of the structuring element depends upon the
nature of analysis and the input given image. The values of the intensities used for A and B
depend upon the input texture image and the nature of the analysis. Various schemes for allotting
the intensity values to set A and B lead to different aura patterns. In general, the structuring
element is denoted by N = { Ns, s S }, where, Ns is one of the elements (single pixel intensity)
taken from the set S or its subset A or subset B. The elements around which the aura is to be
determined are taken from the set A. The intensity values of the surrounding elements are taken
from the set B. The set of matching patterns in the given texture image as per the structuring
element is called aura matrix. In another words, the aura of A w.r.t. B is given by
The total number of aura patterns obtained for a combination of pixel intensity values taken from
set A and set B according to the structuring element is called the aura measure for that
combination and is given by
Fig. 1. Example of an aura on a binary lattice with four nearest neighbors system. (a) A sample
binary lattice S, where the subset A is the set of all 1s and B the set of all 0s. (b)The structuring
element of the neighborhood system (c) The set of shaded sites is the aura set of A with respect to
B.
98
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
PA and PB are the patterns formed by taking elements from set A and set B, respectively according
to the structuring element. Pt is the pattern formed in the input texture at the position where the
structuring element is currently placed. PAB is the pattern formed from the elements set A and set
B. Hence the aura measure for a combination of PA and PB at a particular site in the input texture
is equal to total number of matchings of all possible patterns formed from set A & set B with
respect to the pattern formed below the current position of the structuring element. It should be
noted that the matching of patterns is to be computed only at the sites where the intensity values
in the input texture are identical to the elements in set A.
The matrix of all the patterns which can be formed for different combinations of the intensity
values taken from set A and B with respect to the given structuring element is called aura matrix.
For simplicity the image textures are analyzed using their gray level representations. GLAM
(Gray-Level Aura Matrix) defines the aura measure for a typical the structuring element and the
texture image shown in Fig. 1 between Si and Sj using the relation
Where Si and Sj are the ith and jth gray levels in the range {0 i , j 1}
The standard mathematical formulation of aura described in this section is limited in application
because of the need of exact matching of the gray values of structuring elements in with the
image under processing. We have slightly relaxed some of the constraints of the standard aura
and the modified algorithm is explained in the next section.
V. METHODOLOGY
It is clear from the literature survey that limited work has been carried out using the standard
concepts of aura based transformation. Our previous investigations have shown that the power of
aura transformation may be enhanced by modifying some of the concepts of aura technique [2125]. In the present research work, the basic concepts of aura are further modified to widen its
application area. Here, modified aura concept has been used for investigating and enhancing the
quality of MRI scans for better medical diagnosis. The modified aura based algorithm is shown in
Fig. 2 in the form of a flow chart.
In preprocessing, the input MRI scan is converted to gray scale and its size is modified to reduce
the number of computations. The reduction depends upon the expected size and texture of the
abnormal region in the scan.
Different types of abnormalities are introduced in the preprocessed scans for investigating the
effect of aura on these abnormalities and the neighborhood structures of the brain.The
abnormalities like Alzheimer, calcium migraine and other growth of tissues called tumor were
manually introduced. A neighborhood element is defined in the form of a 3x3 matrix.
The values of the elements of this matrix are estimated on the basis of gray scale values of the
given MRI scan. The input image is processed using this structuring element by traversing it pixel
by pixel on the whole image. At every placement, the differences of the gray scale values of the
neighborhood elements and the corresponding pixels below it are computed. Depending upon the
difference threshold Td, the 3x3 matrix of the differences is converted to zeros and ones. If the
difference is less than Td, the corresponding element is mar ked as one otherwise, zero in the
99
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
difference matrix. If the total number of ones in the difference matrix is more than a threshold
called matching threshold Tm, the pixel corresponding to the central element of the neighborhood
element is marked as black, otherwise left unchanged. This process is repeated for the entire input
image. The investigations have been carried out with different values of both the thresholds and
input MRI scan involving various abnormalities. The evaluation of the enhancement in the
processed MRI scan with respect to the input image was carried out using the visual inspection.
VI. RESULTS
The investigations were carried out using two sets of MRI scans. The first set included the images
where the abnormality was introduced manually with varying sizes, shapes, positions and
textures. The second set included the scans having actual abnormalities corresponding to different
subjects and diseases.
100
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
Fig. 3. Input and the corresponding processed MRI scans with different types of abnormalities
and threshold values.
The images in these two sets were processed as per the modified aura methodology described in
the previous section. The results were obtained by modifying the values of the thresholds Td and
Tm. Some of the processed images are shown in Fig. 3. The input images are shown in the column
I. The value of Td for these images was fixed at 30% of the maximum intensity value in the image
under consideration. The corresponding processed images for values of Tm as 3, 7, and 9 are
shown in columns II-IV, respectively.
The visual analysis of the processed images with respect to the input images shows that the
enhancement is better for the moderate values of the matching threshold i.e. around Tm = 7. As
the value of Tm increased beyond 7, the quality of the processed images started deteriorating.
Similarly, it was also observed that when the value of difference threshold Td was fixed in the
range 25 -30 percent of the maximum intensity, better enhancement was achieved.
101
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
VII. CONCLUSION
In this study, investigations were carried out to enhance the quality of the MRI scans involving
different types of abnormalities using modified aura based framework. It was observed that
modified framework with some constrains relaxed, is relatively less computationally expensive,
simple, and promising. The investigations further showed that the processed MRI scans were
enhanced in quality. The enhanced images may be used for predicting the abnormalities in brain
more effectively and accurately. The investigations involving the images obtained from other
medical imaging techniques are in our future plan.
REFERENCES
[1]
L.P.Clarke, R.P. Velthuizen, M.A. Camacho, J.J. Heine, M. Vaidyanathan, L.O. Hall, R.W. Thatcher,
and M.L. Silbiger , MRI segmentation: methods and applications, Magn. Resonance Imag., vol. 13,
no. 3, pp. 343-368, 1995.
[2]
[3]
[4]
V. P. Gladis, P. Rathi and S. Palani, Detection and characterization of brain tumor using
segmentation based on hsom, wavelet packet feature spaces and ann, in Proc. IEEE 3rd Int. Conf.
ICECT, 2011, Kanyakumari, vol. 6, pp. 274-277.
[5]
[6]
A. K. Jain and F. Farrokhnia, Unsupervised texture segmentation using Gabor filters, Putt. Recogn.,
vol. 24, no. 12, pp. 1167-1186, Dec. 1991.
[7]
[8]
G. Mirajkar and B. Barbadekar, Automatic segmentation of brain tumors from MR Images using
undecimated wavelet transform and Gabor wavelets, in Proc. IEEE Trans. ICECS, 2011, Athens, pp.
702-705.
[9]
S. Chen and D. Zhang, Robust image segmentation using FCM with spatialconstraints based on new
kernel induced distance measure, IEEE Systems, Man, and Cybernetics Society, vol. 34, pp. 10971916, 2004
[10] S. Shen, W. Sandham, M. Granat, and A. Sterr, MRI fuzzy segmentation of brain tissue using
neighborhood attraction with neural-network optimization, IEEE Trans. on Inf. Tech. in Biomed.,
vol. 9, pp. 459-467, 2005.
[11] R. Stokking, K. L. Vincken, and M. A. Viergever, Automatic morphology-based brain segmentation
(MBRASE) from MRI-T1 data, NeuroImage, vol. 12, pp. 726-738, 2000.
102
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
[12] J. R. J. Alaniz, V. M. Banuelos, O. Y. Suarez, Data-driven brain MRI segmentation supported on
edge confidence and a priori issue information, IEEE Trans. on Med. Imag., vol. 25, pp. 74-83, 2006
[13] N. Kriegeskorte and R. Goebel, An efficient algorithm for topologically correct segmentation of the
cortical sheet in anatomical MR volumes, NeuroImage, vol. 14, pp. 329- 346, 2001.
[14] M. R. Kaus and S. K. Warfield, A. Nabavi, P. M. Black, F. A. Jolesz, and R. Kikinis, Automated
segmentation of MR images of brain tumors, Radiology, vol. 218, pp. 586-591, 2001.
[15] M. C. Clark, L. O. Hall, D. B. Goldgof, and R. Velthuizen, F. Reed Murtagh, and M. S. Silbiger,
Automatic tumor segmentation using knowledge-based techniques, IEEE Trans. on Med. Imag.,
vol. 17, pp. 187-201, 1999.
[16] M. Prastawa, E. Bullitt, S. Ho, and G. Gerig, A brain tumor segmentation framework based on
outlier detection, Medical Image Analysis, vol. 8, pp. 275-283, 2004.
[17] R. W. Picard and I. Elfadel, Structure of aura and co-occurrence matrices for the Gibbs texture
model, J. Mathematical Imag. and Vision, vol. 2, pp. 5-25, 1992.
[18] R.W. Picard and I. Elfadel, Gibbs random fields, co-occurrences, and texture modeling, IEEE
Trans. Pattern Recognition and Machine Intelligence, vol. 16, no. 1, 1994
[19] X. Qin, Y. H. Yang, Basic gray level aura matrices: theory and its applications to texture synthesis,
in Proc.10th International Conference on Computer Vision, 2005.
[20] X. Qin and Y. H. Yang, Aura 3D textures, IEEE Trans. Vis. Comput. Graphics, vol. 13, no.2, 2007.
[21] P. Abrol, D. Padha, and P. K. Lehana, Application of aura in Identification of textural objects,
PCTE J. Computer Science, vol.3, issue no.2, Jul-Dec, 2008.
[22] P. Abrol, D. Padha, and P. K. Lehana, Effect of distortion on the synthesized textures using aura
based techniques, in Proc. National Conf. Video, Cyber Learning and Electronics, 2008, NITTTR,
Chandigarh.
[23] P. Abrol and D. Padha, Textural visualization: algorithm and analysis, PCTE J. Computer Sciences,
vol.3, issue no.1, pp. 54-58, Jan-June, 2007.
[24] P. Abrol, D. Padha, P.K. Lehana, and I. Gandotra, Transmission of synthesized textures over low
bandwidth networks, in Proc. AICTE sponsored National Conf. next Generation Computing and Inf.
Sys., 2007, MIET, Jammu.
[25] P. Abrol, D. Padha, and P. K. Lehana, Development of a dsp based textural visualization system, in
Proc. UGC sponsored National Seminar on Data Mining Techniques & Applications, Dept. of
Computer Applications, 2007, M. C., Kerala, pp. 23-25.
[26] D. Weishaupt, B. Marincek, and V. D. Koechli, How does MRI work?: an introduction to the physics
and function of magnetic resonance imaging. Springer, 2nd edition, 2003.
[27] Konsylium NZOZ, http://www.tomograf.com.pl/zakresbadan. php (in Polish), retrieved in April 2011.
[28] S. Chaplot, L. M. Patnaik, and N. R. Jagannathan, Classification of magnetic resonance brain images
using wavelets as input to support vector machine and neural network, Biomed. Signal Processing,
vol. 1, pp. 8692, Jun.2006.
103
Signal & Image Processing : An International Journal (SIPIJ) Vol.3, No.1, February 2012
[29] E. A. El-Dahshan, T. Hosny, and A. M. Salem, Hybrid intelligent techniques for MRI brain images
classification, Digital Signal Processing, vol. 20, pp. 433441, 2010.
[30] M. Maitra and A. Chatterjee, Hybrid multiresolution slantlet transform and fuzzy c-means clustering
approach for normal-pathological brain MR image segregation, Med. Eng. Phys., Aug. 2007.
[31] A. Vailaya, M. Figueiredo, A. K. Jain, and H. J. Zhang, Image classification for content-based
indexing, IEEE Trans. Image Processing, vol. 10, no. 1, pp.117-130, 2001.
[32] S. Abbasi and F. Mokhtarian, Affine-similar shape retrieval: application to multiview 3-D object
recognition, IEEE Trans. Image Processing, vol. 10, no. 1, pp.131-139, 2001.
[33] Y. Deng, B. S. Manjunath, C. Kenney, M. S. Moore, and H. Shin, An efficient color representation
for image retrieval, IEEE Trans. Image Processing, vol. 10, no. 1, pp.140-147, 2001.
[34] I. Epifanio and G. Ayala, A random set view of texture classification, IEEE Trans. Image
Processing, vol. 11, no. 8, pp.859-867, 2002.
[35] I.M. Elfadel and R.W. Picard, Gibbs random fields, cooccurrences, and texture modeling, IEEE
Trans. Pattern Analysis and Machine Intelligence, pp. 24-37, 1994.
104