research article a ct reconstruction algorithm based on non...
TRANSCRIPT
Research ArticleA CT Reconstruction Algorithm Based on Non-AliasingContourlet Transform and Compressive Sensing
Lu-zhen Deng Peng Feng Mian-yi Chen Peng He Quang-sang Vo and Biao Wei
The Key Lab of Optoelectronic Technology and Systems of the Education Ministry of China Chongqing UniversityChongqing 400044 China
Correspondence should be addressed to Peng Feng coe-fpcqueducn and Peng He hepeng vvv163com
Received 11 April 2014 Revised 6 June 2014 Accepted 9 June 2014 Published 30 June 2014
Academic Editor Fenglin Liu
Copyright copy 2014 Lu-zhen Deng et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Compressive sensing (CS) theory has great potential for reconstructing CT images from sparse-views projection data Currentlytotal variation (TV-) based CT reconstruction method is a hot research point in medical CT field which uses the gradient operatoras the sparse representation approach during the iteration process However the images reconstructed by this method often sufferthe smoothing problem to improve the quality of reconstructed images this paper proposed a hybrid reconstruction methodcombining TV and non-aliasing Contourlet transform (NACT) and using the Split-Bregman method to solve the optimizationproblem Finally the simulation results show that the proposed algorithm can reconstruct high-quality CT images from few-viewsprojection using less iteration numbers which is more effective in suppressing noise and artefacts than algebraic reconstructiontechnique (ART) and TV-based reconstruction method
1 Introduction
Since computed tomography (CT) [1] technique was bornin 1973 CT has been widely applied in medical diagnoseindustrial nondestructive detection and so forth In medicalCT field how to reconstruct high-quality CT images fromfew-views or sparse-views data is a significant research prob-lem Recently compressive sensing (CS) [2] theory has beenapplied in CT images reconstruction which makes it possibleto reconstruct high-quality images from few-views data InCS theory CT images can be sparsely represented in anappropriate domain such as gradient transform andWavelettransform and the quality of CT reconstructed images will beimproved by some appropriate sparse representations in CTimages reconstruction
Contourlet transform [3] is proposed by Do and Vetterliin 2002 which is a sparse representation for 2D imageswith some properties such asmultiresolutionmultiscale andmultidirection Contourlet transform can also get importantsmooth contour features of the image with few data butthere is frequency aliasing in Contourlet transform Sharpfrequency localization Contourlet transform [4] is firstlyproposed by Lu and Do in 2006 and Feng et al introduced
a detailed explanation and construction in 2009 which isnamed as non-aliasing Contourlet transform (NACT) [5]NACT which can eliminate the frequency aliasing in Con-tourlet transform is more efficient in capturing geometricalstructure and can represent image sparser than traditionalContourlet transform
To solve the optimization problem in CT images recon-struction based on CS Goldstein and Osher proposed Split-Bregman [6]methodwhich is derived fromBregman [7] iter-ation and can accelerate iteration convergence and producebetter reconstruction results Split-Bregman method uses anintermediate variable to split 119871
1regularization and 119871
2regu-
larization into two equations where 1198712and 119871
1regularization
equation can be solved by steepest descent method andthresholding algorithm respectively Based on Split-Bregmanmethod Vandeghinste et al proposed Split-Bregman-basedsparse-view CT reconstruction approach [8] Furthermorean iterative CT reconstruction is proposed using shearlet-based regularization [9] Chu et al proposed multienergy CTreconstruction based on low rank and sparsity with the Split-Bregmanmethod (MLRSS) [10] Chang et al proposed a few-view reweighted sparsity hunting (FRESH) method for CTimages reconstruction [11]
Hindawi Publishing CorporationComputational and Mathematical Methods in MedicineVolume 2014 Article ID 753615 9 pageshttpdxdoiorg1011552014753615
2 Computational and Mathematical Methods in Medicine
In this paper we propose a CT reconstruction algorithmbased on NACT and compressive sensing which tries toexplore the sparse capability of NACT in order to recon-struct high-quality CT images In the following section theproposed algorithm will be introduced In the third sectionwe will analyze the experimental results and discuss relevantissues In the last section Section 4 we will conclude thepaper
2 Theory and Method
21 CT Reconstruction Theory Based on Compressive SensingTheoretically the mathematical CT model can be expressedas
119860119891 = 119901 (1)
where 119860 is the system matrix 119891 is the original imageand 119901 is the projection data Traditional CT reconstructionalgorithms such as filtered backprojection (FBP) [12] andalgebraic reconstruction technique (ART) [13] cannot recon-struct high quality CT images with the sparse sampling orlimited projection data
In 2006 Candes and Donoho put forward the CS theorywhich makes it possible to get high quality CT images withsparse projection data The main idea of CS is that a signalcan be reconstructed with far less sampled frequency thanrequired by conventional Nyquist-Shannon sampling fre-quency if the image has a sparsecompressible representationin a transform domain
Compressive sensing theory can be expressed by thefollowing equation
min 100381710038171003817100381711991010038171003817100381710038170
st 119901 = 119860119891 = 119860Φ119867119910 (2)
where Φ is a orthogonal transform Φ119867 is the correspondinginverse transform and 119891 is the CT image to be reconstructedand has a special relationship withΦ119867 that is 119891 = Φ
119867119910 119901 is
the projection data of 119891 through matrix 119860Inspired by CS theory Sidky et al proposed a total
variation (TV-) based CT reconstruction algorithm usinggradient operator as the sparse representation [14] in whichTV is defined as follows
10038171003817100381710038171198911003817100381710038171003817TV = sum
1003816100381610038161003816nabla1198911003816100381610038161003816 = sum
119904119905
radic(nabla119909119891)2
+ (nabla119910119891)2
= sum
119904119905
radic(119891119904119905minus 119891119904minus1119905
)2
+ (119891119904119905minus 119891119904119905minus1
)2
(3)
where nabla119891 represents gradient operator of an image 119891
22 Non-Aliasing Contourlet Transform The traditionalsparse representation such as gradient operator and Wavelettransform [15] cannot get ideal sparse representation of CTimages In 2002 Ni et al proposed Contourlet transform [16]which can utilize intrinsic structure information of imageto represent images more efficiently compared with Wavelettransform However suffering from frequency aliasing Con-tourlet transform does not show good performance in imagedenoising fusion and enhancement In order to solve this
problem a new multiscale analysis method named non-aliasing Contourlet transform (NACT) was proposed NACTconsists of non-aliasing pyramidal filter banks (NPFB) anddirectional filter banks (DFB) NPFB contains two differentfilter banks 119871
0(120596) 119863
0(120596) and 119871
1(120596) 119863
1(120596) 119871
0(120596) and 119871
1(120596)
mean low-pass filters 1198630(120596) and 119863
1(120596) mean high-pass
filters The relationships of two different filter banks are asfollows
1198632
1(120596) +
1198712
1(120596)
4= 1
1198632
0(120596) + 119871
2
0(120596) = 1
(4)
We assume that 1205961199010
and 1205961199040
represent pass-band fre-quency and stop-band frequency of 119871
0(120596) respectively
Accordingly 1205961199011
and 1205961199041
represent pass-band frequencyand stop-band frequency of 119871
1(120596) respectively In order to
eliminate frequency aliasing the filter parameters shouldmeet (1) 120596
1199041lt 1205872 (2) (120596
1199010+ 1205961199040)2 = 1205872 and (120596
1199011+
1205961199041)2 = 1205874 (3) 120596
1199040le 120587 minus 119886 and 120596
1199041le (120587 minus 119886)2 where 119886
is the maximum width of mixing ingredients in DFB [3]As a sparse representation approach NACT integrate
NPFB and DFB which can decompose image into multidi-rection and multiresolution NPFB decomposes image intoan approximation subband and several detail subbands withdifferent resolutions DFB decomposes the detail subbandsinto directional subbandsThe process of decompositionwith3 levels is shown in Figure 1Wewill use ldquo9-7rdquo filter and ldquopkvardquodirectional filter bank [17] in the study
23 Split Bregman Method In CS theory 1198710norm is the
most ideal regularization norm but it is difficult to solveequations and easily interfered by noise in CT reconstructionso 1198710norm is commonly replaced by 119871
1norm Then the
reconstruction problem depicted by (2) can be converted into
min 100381710038171003817100381711991010038171003817100381710038170
st 119901 = 119860119891 = 119860Φ119867119910 (5a)
min 1003817100381710038171003817Φ11989110038171003817100381710038171
st 119901 = 119860119891 = 119860Φ119867119910 (5b)
where 119910 = Φ119891 Φ is the sparse transform which is normallyused as Wavelet transform Curvelet transform gradientoperator and so forth
Furthermore (5b) can be converted into
119891 = arg min119891
1003817100381710038171003817Φ11989110038171003817100381710038171
+ 1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2 (6)
where 120582 is penalty function weightIn order to solve (6) Goldstein andOsher proposed Split-
Bregman method [6] using an intermediate variable to split1198711regularization and 119871
2regularization into two equations
1198712regularization equation can be solved by gradient descent
method and 1198711regularization equation can be solved by
thresholding algorithm Split-Bregman method contains thefollowing three iteration steps
Step 1
119891119896+1
= arg min119891
1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2+ 120583
10038171003817100381710038171003817119889119896minus Φ119891 minus 119887
11989610038171003817100381710038171003817
2
2 (7)
Computational and Mathematical Methods in Medicine 3
x(n)
L0(120596)
D0(120596)
L1(120596)
L1(120596)
D1(120596)
D1(120596)
darr D2
darr D2
DFB
DFB
DFB
y3(n)
y00(n)simy
07(n)
y10(n)simy
13(n)
y20(n)simy
23(n)
Figure 1 Flowchart of 3 levels of decomposition of NACT
Step 2
119889119896+1
= min119889
1198891 + 12058310038171003817100381710038171003817119889119896minus Φ119891119896+1
minus 11988711989610038171003817100381710038171003817
2
2 (8)
Step 3
119887119896+1
= 119887119896+ (Φ119891
119896+1minus 119889119896+1
) (9)
where 119896 is the Split-Bregman iteration index120583 is convergenceparameter and 119889 and 119887 are intermediate variables with whicheach subproblem can be solved easily
24 Proposed Algorithm According to aforementionedmethods we propose a CT reconstruction algorithm basedon NACT and compressive sensing method which can bedefined as a constrained form (10) or an unconstrained form(11) as follows
min 10038171003817100381710038171198911003817100381710038171003817TV +
1003817100381710038171003817Φ11989110038171003817100381710038171
st 1003817100381710038171003817119860119891 minus 1199011003817100381710038171003817
2
2lt 1205902
(10)
min 10038171003817100381710038171198911003817100381710038171003817TV +
1003817100381710038171003817Φ11989110038171003817100381710038171
+ 1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2 (11)
Applying the Split-Bregman method to (11) we have thefollowing three iteration steps
Step 1
119891119896+1
= arg min119891
1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2+ 120574
10038171003817100381710038171003817119889119896minus nabla119891 minus 119887
11989610038171003817100381710038171003817
2
2
+ 12058310038171003817100381710038171003817119889119896
120593minus Φ119891 minus 119887
119896
120593
10038171003817100381710038171003817
2
2
(12)
Step 2
119889119896+1
= min119889
1198891 + 12057410038171003817100381710038171003817119889119896minus nabla119891119896+1
minus 11988711989610038171003817100381710038171003817
2
2
119889119896+1
120593= min119889120593
10038171003817100381710038171003817119889120593
100381710038171003817100381710038171+ 120583
10038171003817100381710038171003817119889119896
120593minus Φ119891119896+1
minus 119887119896
120593
10038171003817100381710038171003817
2
2
(13)
Step 3
119887119896+1
= 119887119896+ (nabla119891
119896+1minus 119889119896+1
)
119887119896+1
120593= 119887119896
120593+ (Φ119891
119896+1minus 119889119896+1
120593)
(14)
where 120574 is convergence parameter and 119889120593and 119887120593are interme-
diate variables
The steepest descent method is applied to solve (12) Thederivative of (12) is calculated as follows
119892 [119899119898 + 1]
= 2120582119860119879
119898(119860119898119891119898minus 119901119898) minus 2120574nabla
119879(119889119896minus nabla119891119898minus 119887119896)
minus 2120583Φ119879(119889119896
120593minus Φ119891119898minus 119887119896
120593)
119891119898+1
= 119891119898+ 120572119892 [119899119898 + 1]
(15)
where 119899 denotes the iteration index of the steepest descentmethod 119898 = 2 119873data denotes the projection angles 119860
119898
is mth row vector and system matrix 119860 includes 119873data rowvector 119860
119898 Accordingly 119873data row vectors 119901
119898compose the
projection-data vector 119901 120572 is an appropriate step size TheARTmethod is used to get initial image of iteration Equation(13) can be explicitly computed as (16) using the shrinkageoperator as follows
119889119896+1
= shrink(nabla119891119896+1 + 1198871198961
120582)
119889119896+1
120593= shrink(Φ119891119896+1 + 119887
119896
1205931
120583)
(16)
We now describe the iterative steps of the proposed algo-rithm The iteration process contains two loops the outsideloop operate ART and the inside loop solve the optimizationproblem which is constrained by TV and NACTThe outsideloop is labeled by 119899 and the inside loop is labeled by 119896The steps comprising each loop are the DATA-step whichenforces consistency with the projection data the POS-stepwhich ensures a nonnegative imageWe use119891(ART-DATA)[119899119898]
to denote the 119898th DATA-step subiteration with the 119899thiteration and 119891
(ART-POS)[119899] to denote the POS-step with the
119899th iteration in the outside loopWe use119891(NACTTV-DATA)[119896119898]
to denote the 119898th DATA-step subiteration with the 119896thiteration and 119891
(NACTTV-POS)[119896] to denote the POS-step with
the 119896th iteration in the inside loopThe steps of the algorithmare as follows
(A) initialization
119899 = 1 119891(ART-DATA)
[119899 1] = 0 (17)
(B) data projection iteration for119898 = 2 119873data
119891(ART-DATA)
[119899119898] = 119891(ART-DATA)
[119899119898 minus 1]
+ 119860119898
119901119898minus 119860119898sdot 119891(ART-DATA)
[119899119898 minus 1]
119860119898sdot 119860119898
(18)
4 Computational and Mathematical Methods in Medicine
Figure 2 Head phantom
(C) positivity constraint
(119891119894119895)(ART-POS)
[119899]
= (119891119894119895)(ART-DATA)
[119899119873data] (119891119894119895)(ART-DATA)
[119899119873data] ge 0
0 (119891119894119895)(ART-DATA)
[119899119873data] lt 0
(19)
(D) initialization of Split-Bregman
119896 = 1
119889 (119899) =10038171003817100381710038171003817119891(ART-DATA)
[119899 1] minus 119891(ART-POS)
[119899]100381710038171003817100381710038172
119891(NACTTV-DATA)
[119896 1] = 119891(ART-POS)
[119899]
119889119896
119909= nabla119909119891(ART-POS)
[119899]
119889119896
119910= nabla119910119891(ART-POS)
[119899]
119889119896
120593= Φ119891(ART-POS)
[119899]
119887119896
119909= 119887119896
119910= 119887119896
120593= 0
(20)
(E) iteration for119898 = 2 119873data
119889119901= 119860119898119891(NACTTV-DATA)
[119896119898 minus 1] minus 119901119898minus1
119892 [119896119898 minus 1]
= 2120582119860119898119889119901minus 2120574nabla
119879
119909(119889119896
119909minus nabla119909119891 minus 119887119896
119909)
minus 2120574nabla119879
119910(119889119896
119910minus nabla119910119891 minus 119887119896
119910)
minus 2120583Φ119879(119889119896
120593minus Φ119891 minus 119887
119896
120593)10038161003816100381610038161003816119891=119891(NACTTV-DATA)
[119896119898minus1]
119892 [119896119898 minus 1] =119892 [119896119898 minus 1]
1003816100381610038161003816119892 [119896119898 minus 1]1003816100381610038161003816
119891(NACTTV-DATA)
[119896119898]
= 119891(NACTTV-DATA)
[119896119898 minus 1] minus 119886119889 (119899) 119892 [119896119898 minus 1]
(21)
(F) positivity constraint
(119891119894119895)(NACTTV-POS)
[119896 + 1]
= (119891119894119895)(NACTTV-DATA)
[119896119873data] (119891119894119895)(NACTTV-DATA)
[119896119873data] ge 0
0 (119891119894119895)(NACTTV-DATA)
[119896119873data] lt 0
(22)
(G) update 119889119909 119889119910 119889120593 119887119909 119887119910 119887120593 increase 119896 and return to
step (E) until 119896 = 119870NACTTV as follows
119889119896+1
119909= shrink(nabla
119909119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199091
120582)
119887119896+1
119910= shrink(nabla
119910119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199101
120582)
119889119896+1
120593= shrink(Φ119891(NACTTV-POS) [119896 + 1 1] + 119887
119896
1205931
120583)
119887119896+1
119909= 119887119896
119909+ (nabla119909119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119909)
119887119896+1
119910= 119887119896
119910+ (nabla119910119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119910)
119887119896+1
120593= 119887119896
120593+ (Φ119891
(NACTTV-POS)[119896 + 1 1] minus 119889
119896+1
120593)
(23)
(H) initialize next loop
119891(ART-DATA)
[119899 + 1 1] = 119891(NACTTV-POS)
[119870NACTTV 1] (24)
increase 119899 and return to step (B) The iteration is stoppedwhen 119860119891 minus 119901
2
2lt 1205902 In our study we selected 120582 = 1000
120574 = 30 120583 = 30 119886 = 02 and 119870NACTTV = 10 which canstrike a good balance in the steepest descent and generategood reconstruction results in the experiments
3 Experimental Results
31 The Image Quality Evaluation This paper uses the rootmean square errors (RMSE) and universal quality index(UQI) [18] to evaluate the quality of the reconstructed images
RMSE is the most widely applied way to evaluate imagequality and RMSE is defined as
RMSE = radic
1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891119877
119894119895)2
(25)
where 119891119894119895
is the pixel value of original image and 119891119877
119894119895is the
pixel value of reconstructed imageWang and Bovic proposed UQI mode which evaluates
images distortion problem including correlation distortionbrightness distortion and contrast distortion The value of
Computational and Mathematical Methods in Medicine 5
(a) (b) (c)
(d) (e) (f)
Figure 3The reconstructed images using three different reconstruction algorithms from the noise-free and noisy data Top row is for noise-free data and bottom row is for noisy data (a) and (d) are reconstructed by ART (b) and (e) are reconstructed by ART-TV and (c) and (f)are reconstructed by SpBr-NACT method
UQI is between minus1 and 1 When the reconstructed image isthe same as the original image the value of UQI is 1 UQI isdefined as
UQI =4120590119891119891119877119891 times 119891
119877
(1205902
119891+ 1205902
119891119877) [(119891)
2
+ (119891119877
)
2
]
(26)
where
119891 =1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119894119895
119891119877
=1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119877
119894119895
1205902
119891=
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891)2
1205902
119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119877
119894119895minus 119891119877
)
2
120590119891119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891) (119891
119877
119894119895minus 119891119877
)
(27)
32 Numerical Simulation In this section a head phantomas shown in Figure 2 is used to reconstruct and compareby 3 different methods ART ART-TV and our proposedalgorithm (SpBr-NACT)The size of phantom image is 200 times200 We assume that the CT system was viewed as in a typicalpencil-beam geometry and the scanning range was from 1∘to 360∘ with a 120579 angular increment projection angles can beindicated as
120579119894= 1 + 360 times
(119894 minus 1)
119873view 119894 = 1 2
119873view2
120579119894= 182 + 360 times
(119894 minus 119873view2)
119873view
119894 =119873view2
119873view2
+ 1 119873view
(28)
In the simulation we reconstruct the head phantomfrom noise-free and noisy projection data To obtain noisyprojection data we add 10 dB Gaussian noise into noise-free projection data Projection number 119873view is 60 anditeration numbers for all reconstruction algorithms are 50The reconstructed images are shown in Figure 3 and theprofile of line 140 in different reconstructed images is plottedin Figure 4
6 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART
Noisy-free
(a)
OriginalART
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(b)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART-TV
Noisy-free
(c)
OriginalART-TV
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(d)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalSpBr-NACT
Noisy-free
(e)
SpBr-NACT
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
Original
(f)
Figure 4 The profile of line 140 in different reconstructed images Left column is for noise-free date and right column is for noisy data (a)and (b) are for ART method (c) and (d) are for ART-TV method (e) and (f) are for SpBr-NACT method
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
2 Computational and Mathematical Methods in Medicine
In this paper we propose a CT reconstruction algorithmbased on NACT and compressive sensing which tries toexplore the sparse capability of NACT in order to recon-struct high-quality CT images In the following section theproposed algorithm will be introduced In the third sectionwe will analyze the experimental results and discuss relevantissues In the last section Section 4 we will conclude thepaper
2 Theory and Method
21 CT Reconstruction Theory Based on Compressive SensingTheoretically the mathematical CT model can be expressedas
119860119891 = 119901 (1)
where 119860 is the system matrix 119891 is the original imageand 119901 is the projection data Traditional CT reconstructionalgorithms such as filtered backprojection (FBP) [12] andalgebraic reconstruction technique (ART) [13] cannot recon-struct high quality CT images with the sparse sampling orlimited projection data
In 2006 Candes and Donoho put forward the CS theorywhich makes it possible to get high quality CT images withsparse projection data The main idea of CS is that a signalcan be reconstructed with far less sampled frequency thanrequired by conventional Nyquist-Shannon sampling fre-quency if the image has a sparsecompressible representationin a transform domain
Compressive sensing theory can be expressed by thefollowing equation
min 100381710038171003817100381711991010038171003817100381710038170
st 119901 = 119860119891 = 119860Φ119867119910 (2)
where Φ is a orthogonal transform Φ119867 is the correspondinginverse transform and 119891 is the CT image to be reconstructedand has a special relationship withΦ119867 that is 119891 = Φ
119867119910 119901 is
the projection data of 119891 through matrix 119860Inspired by CS theory Sidky et al proposed a total
variation (TV-) based CT reconstruction algorithm usinggradient operator as the sparse representation [14] in whichTV is defined as follows
10038171003817100381710038171198911003817100381710038171003817TV = sum
1003816100381610038161003816nabla1198911003816100381610038161003816 = sum
119904119905
radic(nabla119909119891)2
+ (nabla119910119891)2
= sum
119904119905
radic(119891119904119905minus 119891119904minus1119905
)2
+ (119891119904119905minus 119891119904119905minus1
)2
(3)
where nabla119891 represents gradient operator of an image 119891
22 Non-Aliasing Contourlet Transform The traditionalsparse representation such as gradient operator and Wavelettransform [15] cannot get ideal sparse representation of CTimages In 2002 Ni et al proposed Contourlet transform [16]which can utilize intrinsic structure information of imageto represent images more efficiently compared with Wavelettransform However suffering from frequency aliasing Con-tourlet transform does not show good performance in imagedenoising fusion and enhancement In order to solve this
problem a new multiscale analysis method named non-aliasing Contourlet transform (NACT) was proposed NACTconsists of non-aliasing pyramidal filter banks (NPFB) anddirectional filter banks (DFB) NPFB contains two differentfilter banks 119871
0(120596) 119863
0(120596) and 119871
1(120596) 119863
1(120596) 119871
0(120596) and 119871
1(120596)
mean low-pass filters 1198630(120596) and 119863
1(120596) mean high-pass
filters The relationships of two different filter banks are asfollows
1198632
1(120596) +
1198712
1(120596)
4= 1
1198632
0(120596) + 119871
2
0(120596) = 1
(4)
We assume that 1205961199010
and 1205961199040
represent pass-band fre-quency and stop-band frequency of 119871
0(120596) respectively
Accordingly 1205961199011
and 1205961199041
represent pass-band frequencyand stop-band frequency of 119871
1(120596) respectively In order to
eliminate frequency aliasing the filter parameters shouldmeet (1) 120596
1199041lt 1205872 (2) (120596
1199010+ 1205961199040)2 = 1205872 and (120596
1199011+
1205961199041)2 = 1205874 (3) 120596
1199040le 120587 minus 119886 and 120596
1199041le (120587 minus 119886)2 where 119886
is the maximum width of mixing ingredients in DFB [3]As a sparse representation approach NACT integrate
NPFB and DFB which can decompose image into multidi-rection and multiresolution NPFB decomposes image intoan approximation subband and several detail subbands withdifferent resolutions DFB decomposes the detail subbandsinto directional subbandsThe process of decompositionwith3 levels is shown in Figure 1Wewill use ldquo9-7rdquo filter and ldquopkvardquodirectional filter bank [17] in the study
23 Split Bregman Method In CS theory 1198710norm is the
most ideal regularization norm but it is difficult to solveequations and easily interfered by noise in CT reconstructionso 1198710norm is commonly replaced by 119871
1norm Then the
reconstruction problem depicted by (2) can be converted into
min 100381710038171003817100381711991010038171003817100381710038170
st 119901 = 119860119891 = 119860Φ119867119910 (5a)
min 1003817100381710038171003817Φ11989110038171003817100381710038171
st 119901 = 119860119891 = 119860Φ119867119910 (5b)
where 119910 = Φ119891 Φ is the sparse transform which is normallyused as Wavelet transform Curvelet transform gradientoperator and so forth
Furthermore (5b) can be converted into
119891 = arg min119891
1003817100381710038171003817Φ11989110038171003817100381710038171
+ 1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2 (6)
where 120582 is penalty function weightIn order to solve (6) Goldstein andOsher proposed Split-
Bregman method [6] using an intermediate variable to split1198711regularization and 119871
2regularization into two equations
1198712regularization equation can be solved by gradient descent
method and 1198711regularization equation can be solved by
thresholding algorithm Split-Bregman method contains thefollowing three iteration steps
Step 1
119891119896+1
= arg min119891
1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2+ 120583
10038171003817100381710038171003817119889119896minus Φ119891 minus 119887
11989610038171003817100381710038171003817
2
2 (7)
Computational and Mathematical Methods in Medicine 3
x(n)
L0(120596)
D0(120596)
L1(120596)
L1(120596)
D1(120596)
D1(120596)
darr D2
darr D2
DFB
DFB
DFB
y3(n)
y00(n)simy
07(n)
y10(n)simy
13(n)
y20(n)simy
23(n)
Figure 1 Flowchart of 3 levels of decomposition of NACT
Step 2
119889119896+1
= min119889
1198891 + 12058310038171003817100381710038171003817119889119896minus Φ119891119896+1
minus 11988711989610038171003817100381710038171003817
2
2 (8)
Step 3
119887119896+1
= 119887119896+ (Φ119891
119896+1minus 119889119896+1
) (9)
where 119896 is the Split-Bregman iteration index120583 is convergenceparameter and 119889 and 119887 are intermediate variables with whicheach subproblem can be solved easily
24 Proposed Algorithm According to aforementionedmethods we propose a CT reconstruction algorithm basedon NACT and compressive sensing method which can bedefined as a constrained form (10) or an unconstrained form(11) as follows
min 10038171003817100381710038171198911003817100381710038171003817TV +
1003817100381710038171003817Φ11989110038171003817100381710038171
st 1003817100381710038171003817119860119891 minus 1199011003817100381710038171003817
2
2lt 1205902
(10)
min 10038171003817100381710038171198911003817100381710038171003817TV +
1003817100381710038171003817Φ11989110038171003817100381710038171
+ 1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2 (11)
Applying the Split-Bregman method to (11) we have thefollowing three iteration steps
Step 1
119891119896+1
= arg min119891
1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2+ 120574
10038171003817100381710038171003817119889119896minus nabla119891 minus 119887
11989610038171003817100381710038171003817
2
2
+ 12058310038171003817100381710038171003817119889119896
120593minus Φ119891 minus 119887
119896
120593
10038171003817100381710038171003817
2
2
(12)
Step 2
119889119896+1
= min119889
1198891 + 12057410038171003817100381710038171003817119889119896minus nabla119891119896+1
minus 11988711989610038171003817100381710038171003817
2
2
119889119896+1
120593= min119889120593
10038171003817100381710038171003817119889120593
100381710038171003817100381710038171+ 120583
10038171003817100381710038171003817119889119896
120593minus Φ119891119896+1
minus 119887119896
120593
10038171003817100381710038171003817
2
2
(13)
Step 3
119887119896+1
= 119887119896+ (nabla119891
119896+1minus 119889119896+1
)
119887119896+1
120593= 119887119896
120593+ (Φ119891
119896+1minus 119889119896+1
120593)
(14)
where 120574 is convergence parameter and 119889120593and 119887120593are interme-
diate variables
The steepest descent method is applied to solve (12) Thederivative of (12) is calculated as follows
119892 [119899119898 + 1]
= 2120582119860119879
119898(119860119898119891119898minus 119901119898) minus 2120574nabla
119879(119889119896minus nabla119891119898minus 119887119896)
minus 2120583Φ119879(119889119896
120593minus Φ119891119898minus 119887119896
120593)
119891119898+1
= 119891119898+ 120572119892 [119899119898 + 1]
(15)
where 119899 denotes the iteration index of the steepest descentmethod 119898 = 2 119873data denotes the projection angles 119860
119898
is mth row vector and system matrix 119860 includes 119873data rowvector 119860
119898 Accordingly 119873data row vectors 119901
119898compose the
projection-data vector 119901 120572 is an appropriate step size TheARTmethod is used to get initial image of iteration Equation(13) can be explicitly computed as (16) using the shrinkageoperator as follows
119889119896+1
= shrink(nabla119891119896+1 + 1198871198961
120582)
119889119896+1
120593= shrink(Φ119891119896+1 + 119887
119896
1205931
120583)
(16)
We now describe the iterative steps of the proposed algo-rithm The iteration process contains two loops the outsideloop operate ART and the inside loop solve the optimizationproblem which is constrained by TV and NACTThe outsideloop is labeled by 119899 and the inside loop is labeled by 119896The steps comprising each loop are the DATA-step whichenforces consistency with the projection data the POS-stepwhich ensures a nonnegative imageWe use119891(ART-DATA)[119899119898]
to denote the 119898th DATA-step subiteration with the 119899thiteration and 119891
(ART-POS)[119899] to denote the POS-step with the
119899th iteration in the outside loopWe use119891(NACTTV-DATA)[119896119898]
to denote the 119898th DATA-step subiteration with the 119896thiteration and 119891
(NACTTV-POS)[119896] to denote the POS-step with
the 119896th iteration in the inside loopThe steps of the algorithmare as follows
(A) initialization
119899 = 1 119891(ART-DATA)
[119899 1] = 0 (17)
(B) data projection iteration for119898 = 2 119873data
119891(ART-DATA)
[119899119898] = 119891(ART-DATA)
[119899119898 minus 1]
+ 119860119898
119901119898minus 119860119898sdot 119891(ART-DATA)
[119899119898 minus 1]
119860119898sdot 119860119898
(18)
4 Computational and Mathematical Methods in Medicine
Figure 2 Head phantom
(C) positivity constraint
(119891119894119895)(ART-POS)
[119899]
= (119891119894119895)(ART-DATA)
[119899119873data] (119891119894119895)(ART-DATA)
[119899119873data] ge 0
0 (119891119894119895)(ART-DATA)
[119899119873data] lt 0
(19)
(D) initialization of Split-Bregman
119896 = 1
119889 (119899) =10038171003817100381710038171003817119891(ART-DATA)
[119899 1] minus 119891(ART-POS)
[119899]100381710038171003817100381710038172
119891(NACTTV-DATA)
[119896 1] = 119891(ART-POS)
[119899]
119889119896
119909= nabla119909119891(ART-POS)
[119899]
119889119896
119910= nabla119910119891(ART-POS)
[119899]
119889119896
120593= Φ119891(ART-POS)
[119899]
119887119896
119909= 119887119896
119910= 119887119896
120593= 0
(20)
(E) iteration for119898 = 2 119873data
119889119901= 119860119898119891(NACTTV-DATA)
[119896119898 minus 1] minus 119901119898minus1
119892 [119896119898 minus 1]
= 2120582119860119898119889119901minus 2120574nabla
119879
119909(119889119896
119909minus nabla119909119891 minus 119887119896
119909)
minus 2120574nabla119879
119910(119889119896
119910minus nabla119910119891 minus 119887119896
119910)
minus 2120583Φ119879(119889119896
120593minus Φ119891 minus 119887
119896
120593)10038161003816100381610038161003816119891=119891(NACTTV-DATA)
[119896119898minus1]
119892 [119896119898 minus 1] =119892 [119896119898 minus 1]
1003816100381610038161003816119892 [119896119898 minus 1]1003816100381610038161003816
119891(NACTTV-DATA)
[119896119898]
= 119891(NACTTV-DATA)
[119896119898 minus 1] minus 119886119889 (119899) 119892 [119896119898 minus 1]
(21)
(F) positivity constraint
(119891119894119895)(NACTTV-POS)
[119896 + 1]
= (119891119894119895)(NACTTV-DATA)
[119896119873data] (119891119894119895)(NACTTV-DATA)
[119896119873data] ge 0
0 (119891119894119895)(NACTTV-DATA)
[119896119873data] lt 0
(22)
(G) update 119889119909 119889119910 119889120593 119887119909 119887119910 119887120593 increase 119896 and return to
step (E) until 119896 = 119870NACTTV as follows
119889119896+1
119909= shrink(nabla
119909119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199091
120582)
119887119896+1
119910= shrink(nabla
119910119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199101
120582)
119889119896+1
120593= shrink(Φ119891(NACTTV-POS) [119896 + 1 1] + 119887
119896
1205931
120583)
119887119896+1
119909= 119887119896
119909+ (nabla119909119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119909)
119887119896+1
119910= 119887119896
119910+ (nabla119910119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119910)
119887119896+1
120593= 119887119896
120593+ (Φ119891
(NACTTV-POS)[119896 + 1 1] minus 119889
119896+1
120593)
(23)
(H) initialize next loop
119891(ART-DATA)
[119899 + 1 1] = 119891(NACTTV-POS)
[119870NACTTV 1] (24)
increase 119899 and return to step (B) The iteration is stoppedwhen 119860119891 minus 119901
2
2lt 1205902 In our study we selected 120582 = 1000
120574 = 30 120583 = 30 119886 = 02 and 119870NACTTV = 10 which canstrike a good balance in the steepest descent and generategood reconstruction results in the experiments
3 Experimental Results
31 The Image Quality Evaluation This paper uses the rootmean square errors (RMSE) and universal quality index(UQI) [18] to evaluate the quality of the reconstructed images
RMSE is the most widely applied way to evaluate imagequality and RMSE is defined as
RMSE = radic
1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891119877
119894119895)2
(25)
where 119891119894119895
is the pixel value of original image and 119891119877
119894119895is the
pixel value of reconstructed imageWang and Bovic proposed UQI mode which evaluates
images distortion problem including correlation distortionbrightness distortion and contrast distortion The value of
Computational and Mathematical Methods in Medicine 5
(a) (b) (c)
(d) (e) (f)
Figure 3The reconstructed images using three different reconstruction algorithms from the noise-free and noisy data Top row is for noise-free data and bottom row is for noisy data (a) and (d) are reconstructed by ART (b) and (e) are reconstructed by ART-TV and (c) and (f)are reconstructed by SpBr-NACT method
UQI is between minus1 and 1 When the reconstructed image isthe same as the original image the value of UQI is 1 UQI isdefined as
UQI =4120590119891119891119877119891 times 119891
119877
(1205902
119891+ 1205902
119891119877) [(119891)
2
+ (119891119877
)
2
]
(26)
where
119891 =1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119894119895
119891119877
=1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119877
119894119895
1205902
119891=
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891)2
1205902
119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119877
119894119895minus 119891119877
)
2
120590119891119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891) (119891
119877
119894119895minus 119891119877
)
(27)
32 Numerical Simulation In this section a head phantomas shown in Figure 2 is used to reconstruct and compareby 3 different methods ART ART-TV and our proposedalgorithm (SpBr-NACT)The size of phantom image is 200 times200 We assume that the CT system was viewed as in a typicalpencil-beam geometry and the scanning range was from 1∘to 360∘ with a 120579 angular increment projection angles can beindicated as
120579119894= 1 + 360 times
(119894 minus 1)
119873view 119894 = 1 2
119873view2
120579119894= 182 + 360 times
(119894 minus 119873view2)
119873view
119894 =119873view2
119873view2
+ 1 119873view
(28)
In the simulation we reconstruct the head phantomfrom noise-free and noisy projection data To obtain noisyprojection data we add 10 dB Gaussian noise into noise-free projection data Projection number 119873view is 60 anditeration numbers for all reconstruction algorithms are 50The reconstructed images are shown in Figure 3 and theprofile of line 140 in different reconstructed images is plottedin Figure 4
6 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART
Noisy-free
(a)
OriginalART
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(b)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART-TV
Noisy-free
(c)
OriginalART-TV
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(d)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalSpBr-NACT
Noisy-free
(e)
SpBr-NACT
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
Original
(f)
Figure 4 The profile of line 140 in different reconstructed images Left column is for noise-free date and right column is for noisy data (a)and (b) are for ART method (c) and (d) are for ART-TV method (e) and (f) are for SpBr-NACT method
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
Computational and Mathematical Methods in Medicine 3
x(n)
L0(120596)
D0(120596)
L1(120596)
L1(120596)
D1(120596)
D1(120596)
darr D2
darr D2
DFB
DFB
DFB
y3(n)
y00(n)simy
07(n)
y10(n)simy
13(n)
y20(n)simy
23(n)
Figure 1 Flowchart of 3 levels of decomposition of NACT
Step 2
119889119896+1
= min119889
1198891 + 12058310038171003817100381710038171003817119889119896minus Φ119891119896+1
minus 11988711989610038171003817100381710038171003817
2
2 (8)
Step 3
119887119896+1
= 119887119896+ (Φ119891
119896+1minus 119889119896+1
) (9)
where 119896 is the Split-Bregman iteration index120583 is convergenceparameter and 119889 and 119887 are intermediate variables with whicheach subproblem can be solved easily
24 Proposed Algorithm According to aforementionedmethods we propose a CT reconstruction algorithm basedon NACT and compressive sensing method which can bedefined as a constrained form (10) or an unconstrained form(11) as follows
min 10038171003817100381710038171198911003817100381710038171003817TV +
1003817100381710038171003817Φ11989110038171003817100381710038171
st 1003817100381710038171003817119860119891 minus 1199011003817100381710038171003817
2
2lt 1205902
(10)
min 10038171003817100381710038171198911003817100381710038171003817TV +
1003817100381710038171003817Φ11989110038171003817100381710038171
+ 1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2 (11)
Applying the Split-Bregman method to (11) we have thefollowing three iteration steps
Step 1
119891119896+1
= arg min119891
1205821003817100381710038171003817119860119891 minus 119901
1003817100381710038171003817
2
2+ 120574
10038171003817100381710038171003817119889119896minus nabla119891 minus 119887
11989610038171003817100381710038171003817
2
2
+ 12058310038171003817100381710038171003817119889119896
120593minus Φ119891 minus 119887
119896
120593
10038171003817100381710038171003817
2
2
(12)
Step 2
119889119896+1
= min119889
1198891 + 12057410038171003817100381710038171003817119889119896minus nabla119891119896+1
minus 11988711989610038171003817100381710038171003817
2
2
119889119896+1
120593= min119889120593
10038171003817100381710038171003817119889120593
100381710038171003817100381710038171+ 120583
10038171003817100381710038171003817119889119896
120593minus Φ119891119896+1
minus 119887119896
120593
10038171003817100381710038171003817
2
2
(13)
Step 3
119887119896+1
= 119887119896+ (nabla119891
119896+1minus 119889119896+1
)
119887119896+1
120593= 119887119896
120593+ (Φ119891
119896+1minus 119889119896+1
120593)
(14)
where 120574 is convergence parameter and 119889120593and 119887120593are interme-
diate variables
The steepest descent method is applied to solve (12) Thederivative of (12) is calculated as follows
119892 [119899119898 + 1]
= 2120582119860119879
119898(119860119898119891119898minus 119901119898) minus 2120574nabla
119879(119889119896minus nabla119891119898minus 119887119896)
minus 2120583Φ119879(119889119896
120593minus Φ119891119898minus 119887119896
120593)
119891119898+1
= 119891119898+ 120572119892 [119899119898 + 1]
(15)
where 119899 denotes the iteration index of the steepest descentmethod 119898 = 2 119873data denotes the projection angles 119860
119898
is mth row vector and system matrix 119860 includes 119873data rowvector 119860
119898 Accordingly 119873data row vectors 119901
119898compose the
projection-data vector 119901 120572 is an appropriate step size TheARTmethod is used to get initial image of iteration Equation(13) can be explicitly computed as (16) using the shrinkageoperator as follows
119889119896+1
= shrink(nabla119891119896+1 + 1198871198961
120582)
119889119896+1
120593= shrink(Φ119891119896+1 + 119887
119896
1205931
120583)
(16)
We now describe the iterative steps of the proposed algo-rithm The iteration process contains two loops the outsideloop operate ART and the inside loop solve the optimizationproblem which is constrained by TV and NACTThe outsideloop is labeled by 119899 and the inside loop is labeled by 119896The steps comprising each loop are the DATA-step whichenforces consistency with the projection data the POS-stepwhich ensures a nonnegative imageWe use119891(ART-DATA)[119899119898]
to denote the 119898th DATA-step subiteration with the 119899thiteration and 119891
(ART-POS)[119899] to denote the POS-step with the
119899th iteration in the outside loopWe use119891(NACTTV-DATA)[119896119898]
to denote the 119898th DATA-step subiteration with the 119896thiteration and 119891
(NACTTV-POS)[119896] to denote the POS-step with
the 119896th iteration in the inside loopThe steps of the algorithmare as follows
(A) initialization
119899 = 1 119891(ART-DATA)
[119899 1] = 0 (17)
(B) data projection iteration for119898 = 2 119873data
119891(ART-DATA)
[119899119898] = 119891(ART-DATA)
[119899119898 minus 1]
+ 119860119898
119901119898minus 119860119898sdot 119891(ART-DATA)
[119899119898 minus 1]
119860119898sdot 119860119898
(18)
4 Computational and Mathematical Methods in Medicine
Figure 2 Head phantom
(C) positivity constraint
(119891119894119895)(ART-POS)
[119899]
= (119891119894119895)(ART-DATA)
[119899119873data] (119891119894119895)(ART-DATA)
[119899119873data] ge 0
0 (119891119894119895)(ART-DATA)
[119899119873data] lt 0
(19)
(D) initialization of Split-Bregman
119896 = 1
119889 (119899) =10038171003817100381710038171003817119891(ART-DATA)
[119899 1] minus 119891(ART-POS)
[119899]100381710038171003817100381710038172
119891(NACTTV-DATA)
[119896 1] = 119891(ART-POS)
[119899]
119889119896
119909= nabla119909119891(ART-POS)
[119899]
119889119896
119910= nabla119910119891(ART-POS)
[119899]
119889119896
120593= Φ119891(ART-POS)
[119899]
119887119896
119909= 119887119896
119910= 119887119896
120593= 0
(20)
(E) iteration for119898 = 2 119873data
119889119901= 119860119898119891(NACTTV-DATA)
[119896119898 minus 1] minus 119901119898minus1
119892 [119896119898 minus 1]
= 2120582119860119898119889119901minus 2120574nabla
119879
119909(119889119896
119909minus nabla119909119891 minus 119887119896
119909)
minus 2120574nabla119879
119910(119889119896
119910minus nabla119910119891 minus 119887119896
119910)
minus 2120583Φ119879(119889119896
120593minus Φ119891 minus 119887
119896
120593)10038161003816100381610038161003816119891=119891(NACTTV-DATA)
[119896119898minus1]
119892 [119896119898 minus 1] =119892 [119896119898 minus 1]
1003816100381610038161003816119892 [119896119898 minus 1]1003816100381610038161003816
119891(NACTTV-DATA)
[119896119898]
= 119891(NACTTV-DATA)
[119896119898 minus 1] minus 119886119889 (119899) 119892 [119896119898 minus 1]
(21)
(F) positivity constraint
(119891119894119895)(NACTTV-POS)
[119896 + 1]
= (119891119894119895)(NACTTV-DATA)
[119896119873data] (119891119894119895)(NACTTV-DATA)
[119896119873data] ge 0
0 (119891119894119895)(NACTTV-DATA)
[119896119873data] lt 0
(22)
(G) update 119889119909 119889119910 119889120593 119887119909 119887119910 119887120593 increase 119896 and return to
step (E) until 119896 = 119870NACTTV as follows
119889119896+1
119909= shrink(nabla
119909119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199091
120582)
119887119896+1
119910= shrink(nabla
119910119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199101
120582)
119889119896+1
120593= shrink(Φ119891(NACTTV-POS) [119896 + 1 1] + 119887
119896
1205931
120583)
119887119896+1
119909= 119887119896
119909+ (nabla119909119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119909)
119887119896+1
119910= 119887119896
119910+ (nabla119910119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119910)
119887119896+1
120593= 119887119896
120593+ (Φ119891
(NACTTV-POS)[119896 + 1 1] minus 119889
119896+1
120593)
(23)
(H) initialize next loop
119891(ART-DATA)
[119899 + 1 1] = 119891(NACTTV-POS)
[119870NACTTV 1] (24)
increase 119899 and return to step (B) The iteration is stoppedwhen 119860119891 minus 119901
2
2lt 1205902 In our study we selected 120582 = 1000
120574 = 30 120583 = 30 119886 = 02 and 119870NACTTV = 10 which canstrike a good balance in the steepest descent and generategood reconstruction results in the experiments
3 Experimental Results
31 The Image Quality Evaluation This paper uses the rootmean square errors (RMSE) and universal quality index(UQI) [18] to evaluate the quality of the reconstructed images
RMSE is the most widely applied way to evaluate imagequality and RMSE is defined as
RMSE = radic
1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891119877
119894119895)2
(25)
where 119891119894119895
is the pixel value of original image and 119891119877
119894119895is the
pixel value of reconstructed imageWang and Bovic proposed UQI mode which evaluates
images distortion problem including correlation distortionbrightness distortion and contrast distortion The value of
Computational and Mathematical Methods in Medicine 5
(a) (b) (c)
(d) (e) (f)
Figure 3The reconstructed images using three different reconstruction algorithms from the noise-free and noisy data Top row is for noise-free data and bottom row is for noisy data (a) and (d) are reconstructed by ART (b) and (e) are reconstructed by ART-TV and (c) and (f)are reconstructed by SpBr-NACT method
UQI is between minus1 and 1 When the reconstructed image isthe same as the original image the value of UQI is 1 UQI isdefined as
UQI =4120590119891119891119877119891 times 119891
119877
(1205902
119891+ 1205902
119891119877) [(119891)
2
+ (119891119877
)
2
]
(26)
where
119891 =1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119894119895
119891119877
=1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119877
119894119895
1205902
119891=
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891)2
1205902
119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119877
119894119895minus 119891119877
)
2
120590119891119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891) (119891
119877
119894119895minus 119891119877
)
(27)
32 Numerical Simulation In this section a head phantomas shown in Figure 2 is used to reconstruct and compareby 3 different methods ART ART-TV and our proposedalgorithm (SpBr-NACT)The size of phantom image is 200 times200 We assume that the CT system was viewed as in a typicalpencil-beam geometry and the scanning range was from 1∘to 360∘ with a 120579 angular increment projection angles can beindicated as
120579119894= 1 + 360 times
(119894 minus 1)
119873view 119894 = 1 2
119873view2
120579119894= 182 + 360 times
(119894 minus 119873view2)
119873view
119894 =119873view2
119873view2
+ 1 119873view
(28)
In the simulation we reconstruct the head phantomfrom noise-free and noisy projection data To obtain noisyprojection data we add 10 dB Gaussian noise into noise-free projection data Projection number 119873view is 60 anditeration numbers for all reconstruction algorithms are 50The reconstructed images are shown in Figure 3 and theprofile of line 140 in different reconstructed images is plottedin Figure 4
6 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART
Noisy-free
(a)
OriginalART
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(b)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART-TV
Noisy-free
(c)
OriginalART-TV
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(d)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalSpBr-NACT
Noisy-free
(e)
SpBr-NACT
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
Original
(f)
Figure 4 The profile of line 140 in different reconstructed images Left column is for noise-free date and right column is for noisy data (a)and (b) are for ART method (c) and (d) are for ART-TV method (e) and (f) are for SpBr-NACT method
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
4 Computational and Mathematical Methods in Medicine
Figure 2 Head phantom
(C) positivity constraint
(119891119894119895)(ART-POS)
[119899]
= (119891119894119895)(ART-DATA)
[119899119873data] (119891119894119895)(ART-DATA)
[119899119873data] ge 0
0 (119891119894119895)(ART-DATA)
[119899119873data] lt 0
(19)
(D) initialization of Split-Bregman
119896 = 1
119889 (119899) =10038171003817100381710038171003817119891(ART-DATA)
[119899 1] minus 119891(ART-POS)
[119899]100381710038171003817100381710038172
119891(NACTTV-DATA)
[119896 1] = 119891(ART-POS)
[119899]
119889119896
119909= nabla119909119891(ART-POS)
[119899]
119889119896
119910= nabla119910119891(ART-POS)
[119899]
119889119896
120593= Φ119891(ART-POS)
[119899]
119887119896
119909= 119887119896
119910= 119887119896
120593= 0
(20)
(E) iteration for119898 = 2 119873data
119889119901= 119860119898119891(NACTTV-DATA)
[119896119898 minus 1] minus 119901119898minus1
119892 [119896119898 minus 1]
= 2120582119860119898119889119901minus 2120574nabla
119879
119909(119889119896
119909minus nabla119909119891 minus 119887119896
119909)
minus 2120574nabla119879
119910(119889119896
119910minus nabla119910119891 minus 119887119896
119910)
minus 2120583Φ119879(119889119896
120593minus Φ119891 minus 119887
119896
120593)10038161003816100381610038161003816119891=119891(NACTTV-DATA)
[119896119898minus1]
119892 [119896119898 minus 1] =119892 [119896119898 minus 1]
1003816100381610038161003816119892 [119896119898 minus 1]1003816100381610038161003816
119891(NACTTV-DATA)
[119896119898]
= 119891(NACTTV-DATA)
[119896119898 minus 1] minus 119886119889 (119899) 119892 [119896119898 minus 1]
(21)
(F) positivity constraint
(119891119894119895)(NACTTV-POS)
[119896 + 1]
= (119891119894119895)(NACTTV-DATA)
[119896119873data] (119891119894119895)(NACTTV-DATA)
[119896119873data] ge 0
0 (119891119894119895)(NACTTV-DATA)
[119896119873data] lt 0
(22)
(G) update 119889119909 119889119910 119889120593 119887119909 119887119910 119887120593 increase 119896 and return to
step (E) until 119896 = 119870NACTTV as follows
119889119896+1
119909= shrink(nabla
119909119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199091
120582)
119887119896+1
119910= shrink(nabla
119910119891(NACTTV-POS)
[119896 + 1 1] + 119887119896
1199101
120582)
119889119896+1
120593= shrink(Φ119891(NACTTV-POS) [119896 + 1 1] + 119887
119896
1205931
120583)
119887119896+1
119909= 119887119896
119909+ (nabla119909119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119909)
119887119896+1
119910= 119887119896
119910+ (nabla119910119891(NACTTV-POS)
[119896 + 1 1] minus 119889119896+1
119910)
119887119896+1
120593= 119887119896
120593+ (Φ119891
(NACTTV-POS)[119896 + 1 1] minus 119889
119896+1
120593)
(23)
(H) initialize next loop
119891(ART-DATA)
[119899 + 1 1] = 119891(NACTTV-POS)
[119870NACTTV 1] (24)
increase 119899 and return to step (B) The iteration is stoppedwhen 119860119891 minus 119901
2
2lt 1205902 In our study we selected 120582 = 1000
120574 = 30 120583 = 30 119886 = 02 and 119870NACTTV = 10 which canstrike a good balance in the steepest descent and generategood reconstruction results in the experiments
3 Experimental Results
31 The Image Quality Evaluation This paper uses the rootmean square errors (RMSE) and universal quality index(UQI) [18] to evaluate the quality of the reconstructed images
RMSE is the most widely applied way to evaluate imagequality and RMSE is defined as
RMSE = radic
1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891119877
119894119895)2
(25)
where 119891119894119895
is the pixel value of original image and 119891119877
119894119895is the
pixel value of reconstructed imageWang and Bovic proposed UQI mode which evaluates
images distortion problem including correlation distortionbrightness distortion and contrast distortion The value of
Computational and Mathematical Methods in Medicine 5
(a) (b) (c)
(d) (e) (f)
Figure 3The reconstructed images using three different reconstruction algorithms from the noise-free and noisy data Top row is for noise-free data and bottom row is for noisy data (a) and (d) are reconstructed by ART (b) and (e) are reconstructed by ART-TV and (c) and (f)are reconstructed by SpBr-NACT method
UQI is between minus1 and 1 When the reconstructed image isthe same as the original image the value of UQI is 1 UQI isdefined as
UQI =4120590119891119891119877119891 times 119891
119877
(1205902
119891+ 1205902
119891119877) [(119891)
2
+ (119891119877
)
2
]
(26)
where
119891 =1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119894119895
119891119877
=1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119877
119894119895
1205902
119891=
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891)2
1205902
119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119877
119894119895minus 119891119877
)
2
120590119891119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891) (119891
119877
119894119895minus 119891119877
)
(27)
32 Numerical Simulation In this section a head phantomas shown in Figure 2 is used to reconstruct and compareby 3 different methods ART ART-TV and our proposedalgorithm (SpBr-NACT)The size of phantom image is 200 times200 We assume that the CT system was viewed as in a typicalpencil-beam geometry and the scanning range was from 1∘to 360∘ with a 120579 angular increment projection angles can beindicated as
120579119894= 1 + 360 times
(119894 minus 1)
119873view 119894 = 1 2
119873view2
120579119894= 182 + 360 times
(119894 minus 119873view2)
119873view
119894 =119873view2
119873view2
+ 1 119873view
(28)
In the simulation we reconstruct the head phantomfrom noise-free and noisy projection data To obtain noisyprojection data we add 10 dB Gaussian noise into noise-free projection data Projection number 119873view is 60 anditeration numbers for all reconstruction algorithms are 50The reconstructed images are shown in Figure 3 and theprofile of line 140 in different reconstructed images is plottedin Figure 4
6 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART
Noisy-free
(a)
OriginalART
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(b)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART-TV
Noisy-free
(c)
OriginalART-TV
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(d)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalSpBr-NACT
Noisy-free
(e)
SpBr-NACT
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
Original
(f)
Figure 4 The profile of line 140 in different reconstructed images Left column is for noise-free date and right column is for noisy data (a)and (b) are for ART method (c) and (d) are for ART-TV method (e) and (f) are for SpBr-NACT method
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
Computational and Mathematical Methods in Medicine 5
(a) (b) (c)
(d) (e) (f)
Figure 3The reconstructed images using three different reconstruction algorithms from the noise-free and noisy data Top row is for noise-free data and bottom row is for noisy data (a) and (d) are reconstructed by ART (b) and (e) are reconstructed by ART-TV and (c) and (f)are reconstructed by SpBr-NACT method
UQI is between minus1 and 1 When the reconstructed image isthe same as the original image the value of UQI is 1 UQI isdefined as
UQI =4120590119891119891119877119891 times 119891
119877
(1205902
119891+ 1205902
119891119877) [(119891)
2
+ (119891119877
)
2
]
(26)
where
119891 =1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119894119895
119891119877
=1
119872 times119873sum
0le119894lt119873
sum
0le119895lt119872
119891119877
119894119895
1205902
119891=
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891)2
1205902
119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119877
119894119895minus 119891119877
)
2
120590119891119891119877 =
1
119872 times119873 minus 1sum
0le119894lt119873
sum
0le119895lt119872
(119891119894119895minus 119891) (119891
119877
119894119895minus 119891119877
)
(27)
32 Numerical Simulation In this section a head phantomas shown in Figure 2 is used to reconstruct and compareby 3 different methods ART ART-TV and our proposedalgorithm (SpBr-NACT)The size of phantom image is 200 times200 We assume that the CT system was viewed as in a typicalpencil-beam geometry and the scanning range was from 1∘to 360∘ with a 120579 angular increment projection angles can beindicated as
120579119894= 1 + 360 times
(119894 minus 1)
119873view 119894 = 1 2
119873view2
120579119894= 182 + 360 times
(119894 minus 119873view2)
119873view
119894 =119873view2
119873view2
+ 1 119873view
(28)
In the simulation we reconstruct the head phantomfrom noise-free and noisy projection data To obtain noisyprojection data we add 10 dB Gaussian noise into noise-free projection data Projection number 119873view is 60 anditeration numbers for all reconstruction algorithms are 50The reconstructed images are shown in Figure 3 and theprofile of line 140 in different reconstructed images is plottedin Figure 4
6 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART
Noisy-free
(a)
OriginalART
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(b)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART-TV
Noisy-free
(c)
OriginalART-TV
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(d)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalSpBr-NACT
Noisy-free
(e)
SpBr-NACT
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
Original
(f)
Figure 4 The profile of line 140 in different reconstructed images Left column is for noise-free date and right column is for noisy data (a)and (b) are for ART method (c) and (d) are for ART-TV method (e) and (f) are for SpBr-NACT method
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
6 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART
Noisy-free
(a)
OriginalART
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(b)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalART-TV
Noisy-free
(c)
OriginalART-TV
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
(d)
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
OriginalSpBr-NACT
Noisy-free
(e)
SpBr-NACT
0 50 100 150 2000
05
1
15
Pixel number
Gre
y va
lue
Noisy
Original
(f)
Figure 4 The profile of line 140 in different reconstructed images Left column is for noise-free date and right column is for noisy data (a)and (b) are for ART method (c) and (d) are for ART-TV method (e) and (f) are for SpBr-NACT method
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
Computational and Mathematical Methods in Medicine 7
0 50 100
002
004
006
008
01
012
Iteration numbers
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
0 50 100Iteration numbers
Noisy-free
ARTART-TVSpBr-NACT
09
092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100002
004
006
008
01
012
Iteration numbers
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
Iteration numbers0 50 100
09
092
094
096
098
1U
QI
Noisy
(d)
Figure 5 The relationship of RMSE and UQI with respect to iteration number (a) and (b) are the RMSE and UQI of reconstructed imagesfrom noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructed imagesfrom noisy data with ART ART-TV and SpBr-NACT method respectively
From Figures 3 and 4 we can see that the reconstructedimages using ART and ART-TV methods contain a lot ofnoise and artifacts while the reconstructed images usingSpBr-NACTmethod contain less noise and artifacts and haveclearer edges
Table 1 lists all the RMSE andUQI calculated from recon-structed images It is obvious that the RMSE of reconstructedimages using SpBr-NACT method is much smaller than thatof reconstructed images using ART and ART-TV methodsthe UQI is much bigger Thus SpBr-NACT method canreconstruct higher quality images
Figure 5 plots the change of RMSE and UQI with respectto iteration number Figure 6 plots the change of RMSE and
UQIwith respect to projection number119873view In both figuresART ART-TV and the proposed SpBr-NACT approach areused to reconstruct images from noise-free and noisy dataThe blue-solid line is for ART the green-dashed line is forART-TV and the red dashed line is SpBr-NACT For Figure 5the projection number is fixed and 119873view is 60 For figure6 the iteration number is fixed and equals 50 From bothFigures it is easy to find that with the increase of projectionnumber or iteration number SpBr-NACT approach canalways get the minimum RMSE and maximum UQI whichmeans that the quality of reconstructed images with SpBr-NACT is better than those with ART and ART-TV And alsowe see from Figure 5 when the iteration number is relatively
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
8 Computational and Mathematical Methods in Medicine
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy-free
ARTART-TVSpBr-NACT
(a)
Projection angles
Noisy-free
ARTART-TVSpBr-NACT
0 50 100 150 200092
094
096
098
1
UQ
I
(b)
ARTART-TVSpBr-NACT
0 50 100 150 2000
002
004
006
008
01
012
Projection angles
RMSE
Noisy
(c)
ARTART-TVSpBr-NACT
0 50 100 150 200Projection angles
092
094
096
098
1U
QI
Noisy
(d)
Figure 6 The relationship of RMSE and UQI with respect to projection number119873view (a) and (b) are the RMSE and UQI of reconstructedimages from noise-free data with ART ART-TV and SpBr-NACT method respectively (c) and (d) are the RMSE and UQI of reconstructedimages from noisy data with ART ART-TV and SpBr-NACT method respectively
Table 1 RMSE and UQI of reconstructed images using three different algorithms
RMSE UQIMethods ART ART-TV SpBr-NACT ART ART-TV SpBr-NACTNoisy-free 00502 00321 00196 09869 09947 09980Noisy 00554 00406 00318 09839 09914 09948
small 3 methods that almost have the same RMSE and UQIwhich implies that our proposed method has no advantage ifthe iteration step does not converge
4 Conclusion
In this study we proposed a CT reconstruction algorithmbased on NACT and compressive sensing The experimental
results demonstrate that the proposed method can recon-struct high-quality images from few-views data and has apotential for reducing the radiation dose in clinical appli-cation In the further research we will try to explore moredirectional information from NACT so as to improve theperformance of SpBr-NACT algorithm especially when theprojection number is far more below what we setup in thecurrent experiment
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
Computational and Mathematical Methods in Medicine 9
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported in part by the National NaturalScience Foundation of China (no 61201346) and the Fun-damental Research Funds for the Central Universities (no106112013CDJZR120020 and no CDJZR14125501)
References
[1] G Wang H Yu and B de Man ldquoAn outlook on X-ray CTresearch and developmentrdquo Medical Physics vol 35 no 3 pp1051ndash1064 2008
[2] D L Donoho ldquoCompressed sensingrdquo IEEE Transactions onInformation Theory vol 52 no 4 pp 1289ndash1306 2006
[3] M N Do and M Vetterli ldquoThe contourlet transform an effi-cient directional multiresolution image representationrdquo IEEETransactions on Image Processing vol 14 no 12 pp 2091ndash21062005
[4] Y Lu and M N Do ldquoA new contourlet transform with sharpfrequency localizationrdquo in Proceedings of the IEEE InternationalConference on Image Processing (ICIP 06) vol 2 pp 1629ndash1632Atlanta Ga USA October 2006
[5] P Feng B Wei Y J Pan and D L Mi ldquoConstruction of non-aliasing pyramidal transformrdquo Acta Electronica Sinica vol 37no 11 pp 2510ndash2514 2009
[6] T Goldstein and S Osher ldquoThe split Bregman method for 1198711-regularized problemsrdquo SIAM Journal on Imaging Sciences vol2 no 2 pp 323ndash343 2009
[7] L Bregman ldquoThe relaxation method of finding the commonpoints of convex sets and its application to the solution ofproblems in convex optimizationrdquo USSR Computational Math-ematics and Mathematical Physics vol 7 pp 200ndash217 1967
[8] B Vandeghinste B Goossens J de Beenhouwer et al ldquoSplit-Bregman-based sparse-view CT reconstructionrdquo in Proceedingsof the 11th International Meeting on Fully Three-DimensionalImage Reconstruction in Radiology and Nuclear Medicine (Fully3D rsquo11) pp 431ndash434 2011
[9] B Vandeghinste B Goossens R van Holen et al ldquoIterative CTreconstruction using shearlet-based regularizationrdquo inMedicalImaging 2012 Physics of Medical Imaging vol 8313 of Proceed-ings of SPIE p 83133I San Diego Calif USA February 2012
[10] J Chu L Li Z Chen G Wang and H Gao ldquoMulti-energyCT reconstruction based on low rank and sparsity with thesplit-bregman method (MLRSS)rdquo in Proceedings of the IEEENuclear Science Symposium and Medical Imaging ConferenceRecord (NSSMIC rsquo12) pp 2411ndash2414 Anaheim Calif USANovember 2012
[11] M Chang L Li Z Chen Y Xiao L Zhang and G Wang ldquoAfew-view reweighted sparsity hunting (FRESH) method for CTimage reconstructionrdquo Journal of X-Ray Science and Technologyvol 21 no 2 pp 161ndash176 2013
[12] S L Zhang W B Li and G F Tang ldquoStudy on imagereconstruction algorithm of filtered backprojectionrdquo Journal ofXianyang Normal University vol 23 no 4 pp 47ndash49 2008
[13] R Gordon R Bender and G T Herman ldquoAlgebraic recon-struction techniques (ART) for three-dimensional electron
microscopy and X-ray photographyrdquo Journal of TheoreticalBiology vol 29 no 3 pp 471ndash481 1970
[14] E Y Sidky C Kao and X Pan ldquoAccurate image reconstructionfrom few-views and limited-angle data in divergent-beam CTrdquoJournal of X-Ray Science and Technology vol 14 no 2 pp 119ndash139 2006
[15] M Abramowitz and C A Stegun A Wavelet Tour of SignalProcessing Academic Press San Diego Calif USA 3rd edition2008
[16] X Ni H LWang L Chen and JMWang ldquoImage compressedsensing based on sparse representation using Contourlet direc-tional subbandsrdquoApplication Research of Computers vol 30 no6 pp 1889ndash1898 2013
[17] S Boyd andLVandenbergheConvexOptimization CambridgeUniversity Press 2004
[18] Y R Liu Research on Objective Full-Reference Image QualityEvaluation Method Computer Science amp Technology NanjingChina 2010
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom
Submit your manuscripts athttpwwwhindawicom
Stem CellsInternational
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MEDIATORSINFLAMMATION
of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Behavioural Neurology
EndocrinologyInternational Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Disease Markers
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
BioMed Research International
OncologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Oxidative Medicine and Cellular Longevity
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
PPAR Research
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
ObesityJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational and Mathematical Methods in Medicine
OphthalmologyJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Diabetes ResearchJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Research and TreatmentAIDS
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Gastroenterology Research and Practice
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Parkinsonrsquos Disease
Evidence-Based Complementary and Alternative Medicine
Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom