An efficient data level fusion of multimodal medical images by cross scale fusion rule. APB.pdf (PDFy mirror)

0.12 MB·English

International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954

An efficient data level fusion of multimodal medical images
by cross scale fusion rule.

[1] AYUSH DOGRA(CORRESPONDING AUTHOR)
PH.D STUDENT (DEPT OF ECE)
MMU,MULLANA,AMBALA

[2]DR.PARVINDER BHALLA
PROFESSOR(DEPT OF ECE)
MMU, MULLANA,AMBALA

ABSTRACT

Medical image fusion can help the physicians to extract the features that may not
be normally visible in images by different modalities. In this paper, propose an
efficient fusion method based on cross scale fusion rule. The performance of the
propose method can be verified by objective evaluation metrics i.e. QAB/F .

INTRODUCTION & MOTIVATION
Image fusion is the process of integrating two different modalities to form a single
modality. Image fusion is playing a crucial role in medical imaging remote
sensing, computer vision , robotics etc. image fusion can be categorized into three
levels (1) data level (2) attribute level (3) symbol level. Though image fusion can
R S. Publication, [email protected] Page 215
International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954
be used in many areas like remote sensing & astronomy. Applications like remote
sensing & astronomy use multisensory fusion to achieve high spatial and high
spectral resolutions by combining the two modalities , one having higher spatial
resolutions and other having higher spectral resolution. There are numerous
applications that have appeared in medical imaging like simultaneous evaluation of
CT, MRI & PET images. In medical field pixel level is regarded as
computationally efficient. Medical image fusion is hot topic of research now days.
Lot of research has been conducted on medical image fusion in the last decade, but
still there is scope in the coming years. Multisensory fusion have large applications
in military security and surveillance areas. In multiview fusion a set of images of
same scence taken by the same sensor but from different view points is fused to
obtain an image with higher resolution . Beside multi sensor & multi view there are
other types of fusion strategies that are well explained in [5].
LITERATURE SURVEY
Since it is well known that lot of efficient & effective techniques have been
proposed during the last ten years. But it is better to consider the most recent
techniques during the last 4-5 years. In 2011, multi focus image fusion technique
based on bilateral gradient sharpness criterion is proposed by Jing Tian & his co
researchers [1]. In 2012, Jian Tian & Li chen proposed a multi focus image fusion
method based on wavelet statistical sharpness measure [2]. In 2013,fusion
techniques using cross bilateral filter is proposed by B.K. Shreyamsha kumar[3].
But a most revolutionary image fusion method based on cross scale coefficient
selection is proposed by Rui shen and his co researchers in 2013 [4] & claims that
his proposed method is efficient than the existing one.
RESEARCH GAPS &FUTURE SCOPE
In paper [4], Dr. Rui shen and his fellow research workers claims that there is no
formation of artifacts in the fused results. Formation of artifacts can not be fully
avoided but it can be reduced. Amount of formation of artifacts can be calculated
by objective evaluation metrics where as author does not done any such evaluation.
Secondly the author does not evaluate the loss of information from source images
to fused images. These are the mere drawbacks that I would like to investigate in
R S. Publication, [email protected] Page 216
International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954
future research work and like to go for the future enhancement of the cross scale
fusion rule .
PROBLEM FORMULATION

Figure 1-MSD based fusion[4]
First the source images are decomposed to multi scale representations using multi
resolution decomposition to various levels. MSR is low resolution pyramidal
structure contains one approximation level and several detail levels.
Approximation level stores low pass coefficients and DET stores high pass
coefficients. Then a proposed cross scale fusion rule is applied to different
coefficients and finally get the resultant synthesized image by applying inverse
MSD.
OBJECTIVE OF THIS PAPER
.(1) To re -implement the cross scale fusion rule proposed by Dr. Rui shen. and
propose a novel techniques of image fusion.
WORK DONE
We have taken one data set of PET & MRI images. Then decomposed these source
images to 3 levels by MSD(DWT). Then imply the Gaussian & Butterworth band
pass filtering of mid frequencies but prefer the Gaussian band pass filter as it avoid
ringing effects that lead to formation of artifacts. Then activity level measurement
is done by CBA (coefficient based activity) for DET of source images. Then
compute the member ship by using zero gauss member ship function as input and
compute the output the triangular membership function for coefficients selection.
R S. Publication, [email protected] Page 217
International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954
If I = 0,I = 0 then select low pass coefficient i.e. APX. If I ≠ 0,I ≠0 then select
X Y x y
high pass coefficients i.e. DET. So as a conclusion we have to select both high pass
& low pass coefficients. Firstly select the low pass coefficient for fusion. Fused the
different levels of APX’S of first source images to another APX’S of second source
images and get the final fused images.

PET IMAGE MRI IMAGE

MULTIRESOLUTION DECOMPOSITION BY DWT FOR MRI (1ST ,2ND & 3RD
LEVEL)

MULTIRESOLUTION DECOMPOSITION BY DWT FOR PET (1ST,2ND ,& 3RD
LEVEL)
R S. Publication, [email protected] Page 218
International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954

GAUSSIAN BAND PASS FILTERING FOR MRI APX( 1ST ,2ND & 3RD LEVEL )

BUTTERWORTH BAND PASS FILETRING FOR MRI APX(1ST,2ND&3RD
LEVEL)

GAUSSIAN BAND PASS FILTERING FOR PET APX( 1ST ,2ND & 3RD LEVEL )

BUTTERWORTH BAND PASS FILETRING FOR PET APX(1ST,2ND&3RD
LEVEL)
R S. Publication, [email protected] Page 219
International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954
ACTIVITY LEVEL MEASURMENT BY
ACTIVITY LEVEL MEASURMENT BY
CBA FOR DET (PET)
CBA FOR DET (MRI)
CALCULATION OF CORRELATION COFFICIENT OF IMAGE WITH MEDIAN
CALCULATION OF CORRELATION COFFICIENT OF IMAGE WITH MEDIAN
FILTER
FILTER
PET1STLEVEL PET 2NDLEVEL PET 3RDLEVEL MRI1STLEVEL MRI 2NDLEVEL MRI 3RDLEVEL
HORIZONTAL DIAGONAL VERTICAL HORIZONTAL DIAGONAL VERTICAL
-0.2741 0.8759 -0.0865 -0.2688 0.8794 -0.0571
0.6603 0.6656 0.6672 0.7400 0.8132 0.7205
0.9584 0.9636 0.5978 0.2122 0.7765 0.8278

MEMBERSHIP FUNCTION FOR MRI
MEMBERSHIP FUNCTION FOR MRI
,PET (APX-1ST,2NDAND 3RDLEVEL) ,PET (DET-1ST,2NDAND 3RDLEVEL)
• IF Ix =0, Iy= 0 then select low coefficient i.e. APX
• IF Ix =0, Iy= 0 then select low coefficient i.e. APX
• IF Ix not equal to 0 ,Iynot equal to 0, then select high coefficient i.e.
• IF Ix not equal to 0 ,Iynot equal to 0, then select high coefficient i.e. DET
DET

FUSING APX 1STLEVEL MRI TO APX 1ST FUSING APX 2ND LEVEL MRI TO APX FUSING APX 3RD LEVEL MRI TO APX
LEVEL PET 2ND LEVEL PET 3RD LEVEL PET
MRI APX 1STLEVEL PET APX 1STLEVEL FUSED IMAGE • MRI APX 2NDLEVEL PET APX 2ND LEVEL FUSEDIMAGE • MRI APX 3RDLEVEL PET APX 3RDLEVEL FUSEDIMAGE

R S. Publication, [email protected] Page 220
International Journal of Advanced Scientific and Technical Research Issue 4 volume 5, Sep. – Oct. 2014

Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954
OBJECTIVE PERFORMANCE
EVALUATION
LEVELS QAB/F
MRIPET (1STLEVEL) 0.7261
MRIPET (2NDLEVEL) 0.6342
MRIPET (3RDLEVEL) 0.7702

CONCLUSION
QAB/F value is almost close to base paper[4]. But there is no control over the
formation of artifacts. Formation of artifacts induces distortions and noise in the
fused images. It is serious matter to avoid for the upcoming steps.
REFERENCES
[1]. Jing Tian, Li Chen, Lihong Ma, Weiyu Yu,"multi focus image fusion using a
bilateral gradient based sharpness criterion" elsevier,optics communications
vol.264,issue 1,jan 2011.
[2]. Jing Tian , Li Chen,"adaptive multifocus image using a wavelet based
statistical sharpness measure"Elsevier,signal processing,vol. 92,issue 9,September
2012.
[3]. B.K. Shreyamsha Kumar,"image fusion based on pixel significance using cross
bilateral filter "signal,image & video processing,springer, october 2013.
[4] Rui Shen, Irene Cheng, Anup Basu," cross scale coefficient slection for
volumetric medical image fusion,”IEEE,TBE,vol.60,no.4,april 2013.
[5] http://staff.utia.cas.cz/sroubekf/papers/EUSIPCO_07_fusion_tut.pdf

R S. Publication, [email protected] Page 221