Copyright © 2015 American Scientific Publishers
All rights reserved
Printed in the United States of America
RESEARCH ARTICLE
Journal of Medical Imaging and
Health Informatics
Vol. 5, 1–9, 2015
Magnetic Resonance Brain Image Classification via
Stationary Wavelet Transform and Generalized
Eigenvalue Proximal Support Vector Machine
Yudong Zhang
1 2 ∗
, Zhengchao Dong
3
, Aijun Liu
4 5
, Shuihua Wang
1 2 6
,
Genlin Ji
1 2
, Zheng Zhang
7
, and Jiquan Yang
2
1
School of Computer Science and Technology, Nanjing Normal University, Nanjing, Jiangsu 210023, China
2
Jiangsu Key Laboratory of 3D Printing Equipment and Manufacturing, Nanjing, Jiangsu 210042, China
3
Translational Imaging Division and MRI Unit, Columbia University and New York State Psychiatric Institute,
New York, NY 10032, USA
4
School of Economics and Management, Xidian University, Xi’an 710071, China
5
W. P. Carey School of Business, Arizona State University, Tempe, AZ 85287, USA
6
School of Electronic Science and Engineering, Nanjing University, Nanjing, Jiangsu 210046, China
7
Viterbi School of Engineering, University of Southern California, Los Angeles, CA 90089, USA
Background: Automated and accurate classification of MR brain images is of cr ucially importance for medical
analysis and interpretation. We proposed a novel automatic classification system to distinguish abnormal brains
from normal brains in MRI scanning. Methods: Our proposed method used stationary wavelet transform (SWT)
to extract features from MR brain images. Next, principal component analysis (PCA) was harnessed to reduce
the SWT coefficients. Finally, we proposed to use two classifiers, viz., the generalized eigenvalue proximal sup-
port vector machine (GEPSVM), and GEPSVM with RBF kernel. We tested our methods on three benchmark
datasets. Results: The 10 r uns of K-fold cross validation result showed the proposed SWT + PCA + GEPSVM +
RBF method excelled thirteen state-of-the-art methods in terms of classification accuracy. In addition, the
SWT +PCA+ GEPSVM+RBF method achieved accuracy of 100%, 100%, and 99.41% on Dataset-66, Dataset-
160, and Dataset-255, respectively. Conclusion: We proved the effectiveness of both SWT and GEPSVM. The
proposed method may be applied to clinical use.
Keywords: Magnetic Resonance Imaging, Support Vector Machine, Pattern Recognition, Stationary Wavelet
Transfor m, Principle Component Analysis, Radial Basis Function, Classification.
1. INTRODUCTION
Magnetic resonance imaging (MRI) is a low-risk, f ast, non-
invasive imaging technique that produces high quality images of
the anatomical structures of the human body, especially in the
brain, and provides rich information for clinical diagnosis and
biomedical research.
1–3
Soft tissue structures are clearer and more
detailed with MRI than other imaging modalities.
4–6
Numerous
researches are carried out, trying not only to improve the mag-
netic resonance (MR) image quality, but also to seeking novel
methods for easier and quicker pre-clinical diagnosis from MR
images.
The problem arises that existing manual methods of analysis
and interpretation are tedious, time consuming, costly, and irre-
producible, due to the huge amount of imaging data. This neces-
sitates the requirement to develop automatic computer-aided
∗
Author to whom correspondence should be addressed.
diagnosis (CAD) tool. One of the most discriminant features of
a normal brain was its symmetry t hat was obvious in either the
axial or the coronal direction. However, the asymmetry along an
axial MR brain suggested a pathological brain. This symmetry–
asymmetry can be modelled by various image processing tech-
niques, and can be used to classify normal and abnormal brain
MR images.
7 8
In the last decade, various methods were proposed for brain
MR image classification. Chaplot et al.,
9
used the approxima-
tion coefficients obtained by discrete wavelet transform (DWT),
and employed the self-or ganizing map (SOM) neural network
and support vector machine (SVM). Maitra and Chatterjee
10
employed the Slantlet transform, which is an improved version
of DWT. Their feature vector of each image is created by consid-
ering the magnitudes of Slantlet transform outputs corresponding
to six spatial positions chosen according to a specific logic. Then,
they used the common back-propagation neural network (BPNN).
J. Med. Imaging Health Inf. Vol. 5, No. 7, 2015 2156-7018/2015/5/001/009 doi:10.1166/jmihi.2015.1542 1