Medical Image Fusion Method Based on
Lifting Wavelet Transform and Dual-channel PCNN
Yang Yanchun
School of Electronic &
Information Engineering
Lanzhou Jiaotong University
Lanzhou,China
e-mail: yangyanchun102@sina.com
Dang Jianwu
School of Electronic &
Information Engineering
Lanzhou Jiaotong University
Lanzhou,China
e-mail:dangjw@mail.lzjtu.cn
Wang Yangping
School of Electronic &
Information Engineering
Lanzhou Jiaotong University
Lanzhou,China
e-mail:wangyangping@mail.lzjtu.cn
Abstract—In order to further improve the quality of medical
image fusion, the study proposes a medical image fusion method
based on lifting wavelet transform(LWT) and dual-channel pulse
coupled neural network( PCNN). A fusion rule based on region
spatial frequency is adopted in low frequency sub-band
coefficient. Dual-channel PCNN has a simpler network
architecture and better adaptability. It takes less time-consuming
and cuts down computational complexity in the process of large
amoumt of medical images. Dual-channel PCNN fusion rule is
adopted in high frequency sub-band coefficients. The experiment
results show that the proposed method can greatly improve the
quality of fusion image compared with traditional fusion methods
and has less time-consuming with less computational complexity.
Keywords—lifting wavelet transform; dual-channel PCNN;
region spatial frequency; medical image fusion
I. INTRODUCTION
With the development of medical imaging technology,
multi-modality imaging technology has been widely used in
clinical diagnosis and treatment. For example, computed
tomography (CT) is used in tumor and anatomical detection,
magnetic resource imaging (MRI) is used to obtain information
among tissues, positron emission tomography (PET), single
positron emission computed tomography(SPECT) provide
functional and metabolic information. Hence, one can easily
conclude that none of these modalities is able to carry all
relevant information in a single image. Therefore Multi-
modality medical image fusion is required to obtain all possible
relevant information in a single composite image. Medical
image fusion combines the advantages of various modalities
and complementary information to provide more effective
information for medical diagnosis and treatment[1].
The traditional wavelet transform based on convolution as
an effective multi-resolution analysis method, which has been
widely applied[2,3]. But wavelet transform has some defects
such as memory consuming, poor real-time and higher
computational complexity in the processing of large amount of
medical images. To solve this problem medical images are
decomposed by the lifting wavelet transform in the study. The
lifting wavelet transform make calculation simple and faster,
the fusion difficulty will be greatly reduced[4]. Pulse coupled
neural network( PCNN) is a new neural network which is
different from traditional artificial neural network and has great
potential for applications in image fusion research field[5]. The
PCNN is a biologically inspired neural network,which can
retain more details in the field of image fusion, it is more
consistent with the characteristics of human visual system. But
PCNN model has some shortcomings, PCNN is too complex
and has too many parameters,time-consuming is very long.
However dual-channel PCNN has a simpler network
architecture and better adaptability. Using dual-channel PCNN
as a fusion decision can reduce the PCNN model number. Two
external stimuli work in a neuron at the same time, which
makes two images processed in a parallel way and less time-
consuming in the process of image fusion and cuts down
computational complexity as well.
II.
METERIALS AND METHODS
Lifting wavelet transform(LWT)[6]proposed by Sweldens,
is a new wavelet construction method using the lifting scheme
in time domain. The most notable feature in the lifting wavelet
transform is not dependent on the Fourier transform, all
operations are performed in time domain. Its reconstruction can
be implemented by easily adjusting the computation order or
the signs during the process of decomposition so that the
computational complexity of the lifting wavelet transform can
be reduced to half, making the wavelet transform simple and
fast. Therefore, the amount of data which are produced by
managing the two-dimensional image data
can be reduced to three quarters of the original.
The decomposition stage of LWT can be divided into
three parts:split, prediction and update.
The first part is split, which makes the Original
signal
j
into odd samples and even samples, The
decomposition process is expressed as follows:
11
() ( , )
jjj
Fs s d
−−
= ,
1j
−
is low frequency approximation
component,
1j
d
−
is high frequency detail component, ()
j
s
indicates decomposition processing.
The second part is prediction, which uses the even
samples to predict the odd samples by taking advantage of
1179
978-1-4799-4315-9/14/$31.00
c
2014 IEEE