IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO.

9, SEPTEMBER 2014 2367

Iterative Reconstruction for X-Ray Computed

Tomography Using Prior-Image Induced
Nonlocal Regularization
Hua Zhang, Jing Huang, Jianhua Ma∗ , Member, IEEE, Zhaoying Bian, Qianjin Feng, Hongbing Lu,
Zhengrong Liang, Fellow, IEEE, and Wufan Chen∗ , Senior Member, IEEE

Abstract—Repeated X-ray computed tomography (CT) scans I. INTRODUCTION

are often required in several specific applications such as perfusion
-RAY computed tomography (CT) techniques have
imaging, image-guided biopsy needle, image-guided intervention,
and radiotherapy with noticeable benefits. However, the associated
cumulative radiation dose significantly increases as comparison
X been widely explored for specific applications in clinic
including perfusion imaging, image-guided biopsy needle,
with that used in the conventional CT scan, which has raised major image-guided intervention, and radiotherapy with noticeable
concerns in patients. In this study, to realize radiation dose reduc-
tion by reducing the X-ray tube current and exposure time (mAs)
benefits [1]–[4]. Under these situations, repeated tomographic
in repeated CT scans, we propose a prior-image induced nonlo- acquisitions are often prescribed. For instance, in the case of in-
cal (PINL) regularization for statistical iterative reconstruction terventional imaging, a patient may undergo one or more imag-
via the penalized weighted least-squares (PWLS) criteria, which ing studies prior to the current image acquisition, whereas in
we refer to as “PWLS-PINL”. Specifically, the PINL regulariza- other cases such as in daily CBCT examinations for target lo-
tion utilizes the redundant information in the prior image and the
weighted least-squares term considers a data-dependent variance
calization in image-guided radiation therapy (IGRT), repeated
estimation, aiming to improve current low-dose image quality. Sub- scans have become routine procedures. Despite the achieve-
sequently, a modified iterative successive overrelaxation algorithm ments of the repeated CT scans, the cumulative radiation dose
is adopted to optimize the associative objective function. Experi- still significantly increases as comparison with that used in the
mental results on both phantom and patient data show that the conventional CT scans [5], which has raised major concerns in
present PWLS-PINL method can achieve promising gains over the
other existing methods in terms of the noise reduction, low-contrast
patients. Reducing the radiation exposure is one of the major
object detection, and edge detail preservation. efforts in the CT fields [6]–[9]. It is known that lowering the
X-ray tube current and /or shortening the exposure time (mAs)
Index Terms—Penalized weighted least-squares, prior image,
regularization, statistical iterative reconstruction (SIR), X-ray
in CT scan would be a simple and cost-effective method to re-
computed tomography (CT). duce radiation exposure. However, the associated image usually
suffers from serious noise induced artifacts due to the excessive
quantum noise if adequate treatment on the noisy measurement
is not applied.
Manuscript received December 30, 2012; revised September 2, 2013 and Up to now, various techniques have been investigated, in-
October 12, 2013; accepted October 14, 2013. Date of publication Octo-
ber 24, 2013; date of current version August 18, 2014. This work was sup-
cluding optimal scan protocols [7]–[9] and advanced image re-
ported in part by the National Natural Science Foundation of China un- construction algorithms [10]–[14], to reduce radiation dose in
der Grants 81000613, 81101046, 81371544, the National Key Technology CT scans. Among them, the statistical iterative reconstruction
Research and Development Program of the Ministry of Science and Tech-
nology of China under Grant 2011BAI12B03, the Science and Technology
(SIR) methods [12], [13], which take into account the statis-
Program of Guangdong Province of China under Grant 2011A030300005, tical noise properties of the measurement and accommodate
and the 973 Program of China under Grant 2010CB732503. The work the imaging geometry, have shown great potential to reduce the
of Z. Liang wassupported by the NIH/NCI under Grants #CA143111 and
#CA082402. H. Zhang and J. Huang contribute equally to this paper.
quantum noise and artifacts as comparison with the filtered back-
Asterisks indicate corresponding authors. projection (FBP) reconstruction method [15]. A major drawback
H. Zhang, J. Huang, Z. Bian, and Q. Feng are with School of of the SIR methods is the computational burden associated with
Biomedical Engineering, Southern Medical University, Guangzhou 510515,
China (e-mail: xinsier@gmail.com; hjing@fimmu.com; zybian86@gmail.com;
the multiple reprojection and back-projection operation cycles
fengqj99@fimmu.com). through the image domain. However, with the development of
∗ J. Ma is with School of Biomedical Engineering, Southern Medical Uni-
fast computers and dedicated hardware [16], [17], the modified
versity, Guangzhou 510515, China and also with the Department of Radiol-
ogy, State University of New York, Stony Brook, NY 11794 USA (e-mail:
SIR methods have been used recently in advanced CT equip-
jhma@smu.edu.cn, jhma@mil.sunysb.edu). ments [13], [18].
H. Lu is with the Department of Biomedical Engineering, Fourth Military Generally, the SIR methods can be derived from the max-
Medical University, Xi’an 710032, China (e-mail: hblu@fmmu.edu.cn).
Z. Liang is with the Department of Radiology, State University of New York,
imum a posteriori (MAP) estimator given the observed data
Stony Brook, NY 11794 USA (e-mail: jzl@mil.sunysb.edu). or measurement, which usually consist of two terms in the as-
∗ W. Chen is with School of Biomedical Engineering, Southern Medical Uni-
sociative objective function. Specifically, the first term, named
versity, Guangzhou 510515, China (e-mail: chenwf@smu.edu.cn).
Color versions of one or more of the figures in this paper are available online
as the “data-fidelity term,” models the statistical measurement;
at http://ieeexplore.ieee.org. and the second term, named as the “image prior” or “regu-
Digital Object Identifier 10.1109/TBME.2013.2287244 larization term,” penalizes the solution. The data-fidelity term
0018-9294 © 2013 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution
requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
2368 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 9, SEPTEMBER 2014
incorporating an accurate statistical modeling of the measure-
ment is a prerequisite of the SIR algorithms [11]–[13], [18] and
the edge-preserving regularization term plays an important role
in the successful image reconstruction. Usually, the regulariza-
tion term is chosen as a shift-invariant function that penalizes the
differences among local neighboring pixels. These regulariza-
tions through equally smoothing both the noise and edge details
often tend to produce an unfavorable oversmoothing effect [12].
In contrast to the smooth regularization, many edge-preserving
regularizations/priors were proposed in the literature [11], [22].
A typical example is the Huber prior [19]–[21], which replaces
the quadratic penalty function with a nonquadratic penalty func-
tion that increases less rapidly compared with the quadratic
penalty function for sufficiently large arguments. However,
these edge-preserving regularizations/priors mostly rely on the
properties of the local smoothness or edges and do not con-
sider the basic affinity structure information of the desired im-
age, such as the gray levels, edge indicator, dominant direction,
and dominant frequency. To address the aforementioned issues
of the conventional regularizations/priors, an edge-preserving
nonlocal (NL) prior was proposed for CT and positron emis-
sion tomography image reconstructions with encouraging re-
sults [22]–[24] by fully exploiting the density difference infor-
mation and the nonlocal connectivity and continuity information
of the desired image.
With regard to the repeated CT scans, a previously scanned
high-quality diagnostic CT image volume usually contains same
anatomical information with the current scan except for some
anatomical changes due to internal motion or patient weight
change. Generally, the CT scans at different times are often
dealt independently and no systematic attempt has been made
to integrate the valuable patient-specific prior knowledge, i.e.,
the previous scanned data that hold a wealth of prior informa-
tion on the patient-specific anatomy, to promote the subsequent
imaging process. Undoubtedly, performing a low-dose proto-
col in the follow-up CT scan by fully using the previous image
into the current image reconstruction framework, is a promising
and interesting research topic [25], [28], [29], [31], [34]. For
instance, Chen et al. proposed a prior image constrained com-
pressed sensing (PICCS) algorithm [29] to enable view angle
undersampling by integrating a prior image into the reconstruc-
tion procedure. Lauzier extended the PICCS algorithm to the
DR-PICCS algorithm [30] for CT radiation dose reduction us-
ing a statistical model. A weakness of PICCS is that the prior and
current images are taken at the same global geometrical coordi-
nates. This assumption, however, does not necessarily translate
into practical settings like in the IGRT applications. Accurate
registration and voxel consistency may limit the wide use of the
PICCS algorithm. To address this issue, Ma et al. [10] proposed
a low-dose CT image filtering method (named as “ndiNLM” al-
gorithm) by utilizing a high-quality normal-dose scan as priori
information to perform current low-dose CT image restoration
based on nonlocal means criteria [26]. The ndiNLM algorithm
performed well in noise reduction, but actually it is a postpro-
cessing approach without considering the statistical property of
the CT projection data. Another strategy that relaxes image reg-
istration was the dictionary learning-based approach proposed
by Xu et al. [34], which was derived with the consideration of
sparse representation theory via a sparse linear combination of
the patch-based atoms in the dictionary.
In this paper, we aim to lower mAs for radiation reduction
in the case of repeated CT scans and propose a prior-image
induced nonlocal (PINL) regularization for SIR via the penal-
ized weighted least-squares (PWLS) criteria, which we refer
to as “PWLS-PINL.” Specifically, the present PINL regular-
ization utilizes the redundant information in the prior-image
and the weighted least-squares term considers a data-dependent
variance estimation, aiming to improve current low-dose image
quality. Subsequently, a modified iterative successive overrela-
tion algorithm is adopted to optimize the associative objective
function.
II. METHODS AND MATERIALS
A. PWLS Criteria for CT Image Reconstruction
Mathematically, the X-ray CT measurement can be approxi-
mately expressed as a discrete linear system
y = Hμ (1)
where μ denotes the vector of attenuation coefficients, i.e.,
μ = (μ1 , μ2 , . . . , μN ) , and y represents the obtained sinogram
data (projections after system calibration and logarithm trans-
formation), i.e., y = (y1 , y2 , . . . , yM ) , where “ ” denotes the
matrix transpose. The operator H represents the system or pro-
jection matrix with a M × N size. The element of Hij denotes
the length of intersection of projection ray i with pixel j. In
our implementation, the associated element was precalculated
by a fast ray-tracing technique [32] and stored as a file. The
goal of CT image reconstruction is to estimate the attenuation
coefficients μ from the measurement y.
According to the measurement model (1) and the MAP esti-
mation criterion [12], [33], the mathematical formula for PWLS
image reconstruction with a regularization term R(μ) can be ex-
pressed as
 
μ∗ = arg min (y − Hμ) Σ−1 (y − Hμ) + βR(μ) (2)
μ≥0
where Σ is a diagonal matrix with the ith element of σi2 which
is the variance of sinogram data y. β is a hypeparameter that
controls the strength of regularization.
To determine the parameter σi2 in (2), several methods can be
used [12], [34]. In this study, the variance of σi2 was determined
by the mean-variance relationship proposed by Ma et al. [27],
which is written as
 
1 1
2
σi (y) = exp(p̄i ) 1 + exp(p̄i )(σe − 1.25)
2
(3)
I0 I0
where I0 denotes the incident X-ray intensity, p̄i is the mean of
sinogram data y at bin i, and σe2 is the background electronic
noise variance.
B. Overview of the PINL regularization
Traditionally, R(μ) in (2) is designed by a simple weighted
sum of the potential function on the different values of the
ZHANG et al.: ITERATIVE RECONSTRUCTION FOR X-RAY COMPUTED TOMOGRAPHY 2369

neighboring pixels in the image domain [12] and can be de-
scribed as
 
R(μ) = R(μj ) = w(k, j)ϕ(μj − μk ) (4)
j j k ∈S j
where index j runs over all image elements in the image do-
main, Sj represents the local neighborhood of the jth image
pixel in two dimensions, and ϕ denotes a convex and positive
potential function satisfying ϕ(0) = 0. Weight w(k, j) is pos-
itive and symmetric, i.e., w(k, j) ≥ 0 and w(k, j) = w(j, k),
which are usually designed as an inverse proportion of the dis-
tance between the pixels k and j in Sj . These regularizations
are inherently local because they lack global connectivity or
continuity.
In the repeated CT scans, a previously scanned normal-dose
CT image has lower noise level and higher resolution than the
current low-dose CT image, and majority of the anatomical in-
formation are same between the two scans. Thus, in this paper,
to utilize fully the redundancy of information in the prior image,
we propose a prior-image induced nonlocal (PINL) regulariza-
tion with quadratic potential function, which can be written as
 
RPINL (μ) = w(k, j)(μj − μreg
prior,k )
2
(5)
j k ∈N j
where μreg
prior denotes the registered prior image between the
prior image μprior and the current low-dose image μ, Nj denotes
the search window in the image μreg prior . The nonlocal weight
w(k, j) between the prior image and the current objective image
is defined as

w(k, j) = exp(−D(k, j)/h2 )/ exp(−D(k, j)/h2 ) (6)
k ∈N j
D(k, j) = μ(nj ) − μreg 2
prior (nk )2
μ(nj ) = {μl : l ∈ nj }, μreg reg
prior (nk ) = {μprior,l : l ∈ nk }. (8)
Here, weight w(k, j) is a decreasing function of the similarity
between two local neighborhoods nk and nj (named patch-
window) centered at the pixel k in the image domain of μreg prior
and the pixel j in the image domain of μ, respectively. ·22
represents the 2-D Euclidean distance between two similar-
ity patch-windows. μ(nj ) and μreg prior (nk ) denote the vector of
neighborhood pixel values restricted in the patch-windows nj
and nk , respectively, and h is a parameter that controls the decay
of the exponential function.
In summary, the objective function of the PWLS with the
present PINL regularization (i.e., PWLS-PINL) for CT image
reconstruction can be rewritten as
 
μ∗ = arg min (y − Hμ) Σ−1 (y − Hμ) + βRPINL (μ) . (9)
μ≥0
C. Implementation of the PWLS-PINL Method
In the implementation of the present PWLS-PINL method,
the current CT images and the initial high-dose CT images are
first reconstructed by the FBP method, and then a B-spline-
based image registration technique [37] is adopted to generate
the registered prior image from the initial high-dose CT im-
age. The effectiveness of the B-spline-based image registration
algorithm has been extensively validated by many registration
experiments. Based on the transformation matrix, all voxels of
the prior CT images are then transformed into the current CT
images to obtain globally aligned prior images. Similar to our
previous studies in low-dose CT image restoration [10], through
two roughly aligned images, the rich redundant information in
the prior image can be effectively used to induce the PINL regu-
larization for low-dose image reconstruction. Further evaluation
of image registration in our study is presented in Section III-A5.
After we obtained the registered prior image μreg prior , weight
w(k, j) in (6) can be regarded as a function of the current es-
timation μ and the registered prior image μreg prior . Because of
the nonlinearity in calculating weight w(k, j), minimizing the
objective function of (9) is difficult for a closed solution. To
address this problem, based on our previous studies [23], [28],
a binary optimal scheme can be used to minimize the objective
function of (9). Specifically, weight w(k, j) is automatically ad-
justed according to the similarity between the patch-windows
in the current estimation μn (n is the iterative index) and the
registered prior image μreg
prior during each iteration.
In this study, we modified the iterative successive overre-
laxation algorithm described in [12] and [38] to calculate the
solution of (9). In the implementation, the variance σi2 , i.e., the
element of weight matrix in (9), is updated in each iteration
according to (3) for a more accurate estimation of the sino-
gram variance. For simplicity, we call our implementation as
the PWLS-PINL algorithm, and similar to [12], [38], it can be
summarized as follows:
1 Initialization
2 μ= FBP{y}
(7)
3 r̂=y−Hμ̂
4 Σ= diag{σi2 (yi )}
5 Sj = H j Σ−1 Hj , ∀ j.
6 For n = 1, 2, 3, . . . , IterMax,
7 Update {w(k, j)} according to Eq. (6).
8 For each voxel,
9 μ̂old
j := μ̂j

10 αj := Sj + β k ∈N j w(k, j), ∀ j

H j Σ −1 r̂ +S j μ̂ oj l d +β k ∈N j w (k ,j )μ rpergi o r , k
11 μ̂new
j := αj
 
12 μ̂j := max 0, (1 − ω)μold
j + ω μ̂new
j
13 r̂ := r̂ + Hj (μ̂old
j − μ̂j ).
14 End

16 Σ := diag{σi2 ( j Hij μ̂j )}.
17 End
where IterMax denotes the total iterative number, the parameter
ω should satisfy 0 ≤ ω ≤1. The iterative process is terminated
if certain convergence criteria is satisfied for a relatively stable
solution [12], [22]. In this study, the 200th iterated image was
2370 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 9, SEPTEMBER 2014

0.5 s, the slice thickness was 5.0 mm, the tube voltage was
120 kVp, the tube currents are 40 and 400 mA.

E. Evaluation Merits
1) Evaluation by Noise Reduction: The following metrics
were utilized to evaluate the noise reduction on the reconstructed
image from the low-dose sinogram data: 1) signal to noise ratio
(SNR); 2) local signal to noise ratio (lSNR); and 3) relative root
mean square error (rRMSE)
 
Q
m =1 (μm − μ̄)
2
SNR = 10 log10 Q (10)
m =1 (μm − μGS,m )
2
Fig. 1. Illustration of an anthropomorphic torso phantom (a), and (b) a CT
image reconstructed by the FBP method with an optimized Hamming filter from
1
Q
the sinogram data acquired with 100 mAs, 120 kVp. Q m =1 μm
lSNR =
 2 (11)
Q Q
m =1 μm − Q
1 1
selected as the result which corresponds to a stable estimation Q m =1 μm

without visually changing for further iterations. 

Q
m =1 |μm − μQ |2
rRMSE =  Q
(12)
D. Data Acquisition m =1 |μGS,m |2
1) Anthropomorphic Torso Phantom: An anthropomorphic where μ represents the reconstructed image from the low-dose
torso phantom (Radiology Support Devices, Inc., Long Beach, sinogram data, and μGS denotes a golden standard image, m
CA, USA) was used for the experimental data acquisition, as indicates the voxel index in the region of interest (ROI), Q
shown in Fig. 1(a). The phantom was scanned by a clinical CT is the number of voxels in the ROI, and μ̄ denotes the mean
scanner (Siemens SOMATOM Sensation 16 CT) at three expo- voxel value in the ROI. In this study, the golden standard image
sure levels, i.e., 17, 40, 100 mAs. For each exposure level, the is reconstructed by the FBP method with ramp filter from the
tube voltage was set at 120 kVp and the phantom was scanned averaged sinogram data of 150 times repeated scans at 100 mAs,
in a cine mode at a fixed bed position. Fig. 1(b) shows the 120 kVp.
CT image reconstructed by the FBP method with an optimized 2) Evaluation by Image Similarity: To explore the perfor-
Hamming filter from the sinogram data acquired at 100 mAs, mance of the various algorithms at the local detail level, the
120 kVp. The deformed images were simulated by mechani- universal quality index (UQI) [40] was utilized to conduct an
cally performing a cosine transform warped distortion on the ROI-based analysis by evaluating the degree of similarity be-
images reconstructed by the FBP method from the sinogram tween the reconstructed and golden standard images. Given a
data acquired at 100 mAs, 120 kVp. To obtain the registered selected ROI at the corresponding locations in the two images,
prior images, the deformed images were registered to the im- the mean, variance, and covariance of intensities in the ROI can
ages reconstructed by the FBP method from the sinogram data be respectively calculated as
acquired at 17 and 40 mAs, respectively.
2) Numerical XCAT Phantom: To evaluate the influence of 1  1 
Q Q
mismatch between the prior and current images on the per- μ̄ = μm , σ 2 = (μm − μ̄)2 (13)
Q m =1 Q − 1 m =1
formance of the PWLS-PINL and PICCS methods, the XCAT
phantom [39] was used to simulate the normal-dose and low-
1 
Q
dose CT image volumes with the same anatomical structures. 2
μ̄GS = μGS,m , σGS
The associated projection data were simulated using a CT sim- Q m =1
ulator [39] at two exposure levels, i.e., 232 and 23.2 mAs. For
1 
Q
each exposure level, the tube voltage was set at 120 kVp, the time = (μGS,m − μ̄GS )2 (14)
per gantry rotation was 1.0 s, the slice thickness was 3.0 mm. Q − 1 m =1
The system geometry represented a system with 1800 mm
1 
Q
source-to-detector distance, 1510.4 mm center-to-source dis-
cov(μ, μGS ) = (μm − μ̄)(μGS,m − μ̄GS ) (15)
tance, and 0.8448 mm detector pixels (672 radial bins). Recon- Q − 1 m =1
structions were made onto an image volume with the size of
512 × 512 × 100. where μGS denotes a golden standard image, m is the voxel
3) Patient Data: The patient data were obtained with a pa- index, and Q denotes the number of voxels within the ROI.
tient consent for a chest CT study for medical reasons. The Then, the UQI can be calculated as follows:
experimental data were acquired by the Siemens SOMATOM
Sensation 16 CT scanner. The scanning parameters were as fol- 2cov(μ, μGS ) 2μ̄μ̄GS
UQI = 2 . (16)
lows: the pitch was 1.0 mm, the time per gantry rotation was σ 2 + σGS μ̄2 + μ̄2GS
ZHANG et al.: ITERATIVE RECONSTRUCTION FOR X-RAY COMPUTED TOMOGRAPHY 2371

UQI measures the intensity similarity between the two im-

ages, and its value ranges from zero to one. A UQI value closer
to one suggests better similarity to the golden standard image.
3) Evaluation by Contrast-to-Noise Ratio: For quantitative
evaluation of the reconstructed images, we selected two differ-
ent ROIs as indicated by the red squares in Table III for the
calculation of the contrast-to-noise ratio (CNR). The CNR is
defined as
|μ̄ROI − μ̄BG |
CNR =  2 2
(17)
σROI + σBG
where μ̄ROI is the mean of the voxels inside the ROI, and μ̄BG
the mean of the voxels in the background. σROI and σBG are
the standard deviations of the voxel values inside the ROI and
the background region, respectively.

F. Comparison Methods
To validate and evaluate the performance of the present
PWLS-PINL method, the FBP method using ramp filter, the
NL-prior-based PWLS (PWLS-NL) method [23], [24], the
PICCS method [29], [30] and a global dictionary-based sta-
tistical iterative reconstruction (GDSIR) approach [34] were
adopted for comparison. The NL-prior term in the PWLS-NL Fig. 2. Anthropomorphic torso phantom reconstructions by using different
approach is expressed as methods from the sinogram dataset acquired with 17 mAs, 120 kVp. (a) FBP
reconstruction with ramp filter; (b) PWLS-NL reconstruction (N = 21 × 21,
  n = 5 × 5, h = 2.0 × 10−3 , β = 2.5 × 104 ); (c) PICCS reconstruction (α =
RNL (μ) = w(k, j)(μj − μk )2 (18) 0.6, λ = 0.1); (d) image reconstructed by the GDSIR method (γ = 4 × 10−2 )
j k ∈N j with a global dictionary trained from the prior image of Fig. 1(b); and
(e) PWLS-PINL reconstruction using Fig. 1(b) as the prior image (N =
where in weight w(k, j) is defined as 21 × 21, n = 5 × 5, h = 2.0 × 10−3 , β = 2.0 × 104 ). All the images are dis-
  played in a same window.
w(k, j) = exp − μ(nj ) − μ(nk )22 /h2 (19)

μ(nj ) = {μl : l ∈ nj }, μ(nk ) = {μ : l ∈ nk } (20) where D ∈ RN ×K represents the trained dictionary whose col-
umn dk ∈ RN ×1 (k = 1, . . . , K) is a N dimensional vector,
where nj and nk represent the patch-windows, and Nj denotes
αs ∈ RK ×1 is the representation vector with few nonzero en-
the search window in the current low-dose CT image domain.
tries, and  · 0 denotes the l0 -norm. Es ∈ RK ×K is the matrix
The PICCS algorithm under the PWLS framework can be
to extract a patch from the image μ. vs is the Lagrange mul-
formulated as [30]:
tiplier and γ is a hyperparameter. In this study, the global dic-
μ∗ = arg min[αΨ1 (μ − μprior )1 + (1 − α)Ψ2 (μ)1 ] tionary with the size of 64 × 256, (N = 8 × 8, K = 256) was
μ≥0
generated from a normal-dose prior image by using a K-SVD
λ algorithm [35] wherein an orthogonal matching pursuit tech-
+ (y − Hμ) Σ−1 (y − Hμ). (21)
2 nique [36] was used for sparse representation. Additionally, an
where μprior is the prior image, Ψ1 and Ψ2 are the sparsifying alternating minimization scheme described in [34] was adapted
transforms, and α ∈ [0, 1] is a scalar that controls the relative to find the solution of (22).
weights to be allocated to the prior image and compressing
sensing terms. x1 represents the l1 -norm defined as x1 =
 III. RESULTS
i |xi |. In the implementation, the sparsifying transform used
A. Anthropomorphic Torso Phantom Study
for both Ψ1 and Ψ2 is the image spatial gradient norm, which is
also equivalent to the total variation (TV). 1) Noise and Artifacts Suppression: Figs. 2 and 3 show the
The GDSIR is a patch-based approach by extracting the prior results reconstructed by five different methods from the sino-
information via a global dictionary trained from a high-quality gram data acquired at 17 and 40 mAs, respectively. Figs. 2(a)
image [34], and the associative image reconstruction is equiva- and 3(a) show the images reconstructed by the FBP method
lent to solve the following minimization problem: with ramp filter. Serious noise-induced streak artifacts can be
observed. Figs. 2(b) and 3(b) show the images reconstructed by
{μ∗ , α∗ } = arg min (y − Hμ) Σ−1 (y − Hμ)
μ,α the PWLS-NL method. The noise was mostly suppressed, but
 undesired streak artifacts can still be observed in the highly at-
 
+γ Es μ − Dαs 22 + vs αs 0 (22) tenuated regions, partially because of the properties of nonlocal
s s weight calculations for structuring preserving. Figs. 2(c) and
2372 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 9, SEPTEMBER 2014

(a)

Fig. 3. Anthropomorphic torso phantom reconstructions by the four methods (b)

from the sinogram dataset acquired at 40 mAs, 120 kVp. (a) FBP reconstruction
with ramp filter; (b) the PWLS-NL reconstruction (N = 21 × 21, n = 5 × 5,
Fig. 4. Horizontal profiles located at the voxel position y = 227 and x from
h = 2.0 × 10−3 , β = 2.5 × 104 ); (c) PICCS reconstruction using Fig. 1(b) as
425 to 445 as indicated by the red line in Fig. 1(b). (a) are from the different
the prior image (α = 0.6, λ = 0.12); (d) image reconstructed by the GDSIR
reconstructions in Fig. 2 in the case of 17 mAs, 120 kVp; (b) are from the
method (γ = 3 × 10−2 ); and (e) PWLS-PINL reconstruction using Fig. 1(b) as
different reconstructions in Fig. 3 in the case of 40 mAs, 120 kVp.
the prior image (N = 21 × 21, n = 5 × 5, h = 2.0 × 10−3 , β = 2.0 × 104 ).
All the images are displayed in a same window.
TABLE I
IMAGE QUALITY METRICS ON THE THREE ROIS AS INDICATED BY THE
SQUARES IN Fig. 1(b) IN THE CASE OF 17 MAS
3(c) show the images reconstructed by the PICCS method using
Fig. 1(b) as the prior image. The PICCS method can yield rea-
sonable images with sharp edges. Further, the smooth regions
are plagued by some streak artifacts, which may have resulted
from the TV processing. In the context of medical imaging,
this noise might result in some deceptive structures. Figs. 2(d)
and 3(d) show the images reconstructed by the GDSIR method
wherein the global dictionary was trained from the normal-dose
image Fig. 1(b). It can be seen that some distortions occurred
TABLE II
in the case of 17 mAs while satisfactory result appeared with IMAGE QUALITY METRICS ON THE THREE ROIS AS INDICATED BY THE
noticeable noise reduction in the case of 40 mAs. Figs. 2(e) and SQUARES IN Fig. 1(b) IN THE CASE OF 40 MAS
3(e) show the images reconstructed by the present PWLS-PINL
method wherein the prior image is the same as that used in the
PICCS and GDSIR methods. The PWLS-PINL method achieves
more gains over the PICCS and GDSIR methods in terms of
successful noise-induced artifacts suppression and edges detail
preservation. Fig. 4 depicts the associated horizontal profiles
located at the voxel position y = 277 and x from 425 to 445
as indicated by the red line in Fig. 1(b). The profiles from
the present PWLS-PINL method agree much better with those
from the normal-dose scan compared with those from the other GDSIR, and PWLS-PINL, yielded more gains over the FBP re-
methods. constructions from the same sinogram data in terms of the lSNR
To further evaluate quantitatively the present PWLS-PINL and rRMSE measures in the three different ROIs. The results
method, the lSNR and rRMSE of the three different ROIs, as partially demonstrate that the present PWLS-PINL method can
indicated by the squares in Fig. 1(b), were measured in the two achieve noticeable gains over the other methods in terms of noise
cases of 17 and 40 mAs, and are listed in Tables I and II. All and artifacts suppression. However, the PICCS method is supe-
statistical model-based methods, i.e., the PWLS-NL, PICCS, rior to the present PWLS-PINL algorithm in terms of the lSNR
ZHANG et al.: ITERATIVE RECONSTRUCTION FOR X-RAY COMPUTED TOMOGRAPHY
Fig. 5. Zoomed details of the four ROIs in Fig. 2.
Fig. 6. UQI measures on the four ROIs in Fig. 5.
Fig. 7. Zoomed details of the four ROIs in Fig. 3.
and rRMSE measures in the matched region of ROIA due to the
strong same information from the prior image incorporated into
the reconstructed image.
2) UQI Measure: Figs. 5 and 7 show the zoomed details of
the four ROIs in Figs. 2 and 3, respectively. The PWLS-PINL
method achieves significant gains over the other methods in
terms of the resolution preservation. The corresponding UQI
scores are shown in Figs. 6 and 8, respectively. The PWLS-
PINL method yields higher UQI scores (more than 0.98) than
the other methods in both the cases of 17 and 40 mAs.
2373
Fig. 8. UQI measures on the four ROIs in Fig. 7.
TABLE III
CNR MEASURES OF THE IMAGES RECONSTRUCTED BY THE FOUR DIFFERENT
METHODS FROM THE SINOGRAM DATA AT 17 AND 40 MAS
(a) (b)
Fig. 9. Global rRMSE and SNR measures of the present PWLS-PINL method
with respect to the number of iterations from the anthropomorphic torso phantom
reconstructions in the two cases of 17 mAs and 40 mAs. (a) rRMSE versus the
number of iterations; (b) SNR versus the number of iterations.
3) CNR Measure: Table III shows the CNRs of the images
reconstructed by the four different methods from the sinogram
data at 17 and 40 mAs. The ROIs for the CNR measure are
marked by two squares in the figure located at the top left
corner of Table III. Evidently, the present PWLS-PINL method
can achieve significant ability for region contrast identification,
which is very important for low-dose CT image reconstruction.
4) Convergence Analysis of the PWLS-PINL Algorithm: To
validate and analyze the convergence of the present PWLS-
PINL method, global rRMSE and SNR measures on the en-
tire to-be-reconstructed anthropomorphic torso phantom image
were performed. Fig. 9 shows the rRMSE and SNR measures
with respect to the number of iterations in the two cases of 17
and 40 mAs. The results show that the present PWLS-PINL
algorithm can yield a steadily convergence solution in terms of
the global rRMSE and SNR measurements.
2374 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 9, SEPTEMBER 2014
Fig. 10. Influence of registration accuracy on the performance of the
PWLS-PINL method. (a) Elastic deformed FBP image from the original one
reconstructed from the sinogram data acquired at 100 mAs, 120 kVp; (b) Corre-
sponding registered image by using the B-spline-based registration algorithm;
(c) image reconstructed by the PWLS-PINL method from the sinogram data ac-
quired at 17 mAs, 120 kVp using (a) as the prior-image; (d) image reconstructed
by the PWLS-PINL method using (b) as the prior-image from the sinogram data
acquired at 17 mAs, 120 kVp; (e) difference image between (c) and Fig. 2(a);
and (f) difference image between (d) and Fig. 2(a). All the images are displayed
in a same window.
5) Influence of Image Registration Accuracy on the PWLS-
PINL Method: To evaluate the influence of the image registra-
tion accuracy on the performance of the PWLS-PINL method,
we used an elastically deformed FBP image from the original
one reconstructed from the sinogram data acquired at 100 mAs,
120 kVp and the corresponding registered image by the
B-spline-based registration algorithm as the prior images to in-
duce image reconstruction by the PWLS-PINL method. Fig. 10
shows the influence of the registration accuracy on the PWLS-
PINL method in the low-dose image reconstruction from the
sinogram data acquired at 17 mAs, 120 kVp. Figs. 10(a) and (b)
show the elastic deformed FBP image and the corresponding
registered FBP image, respectively. The deformation between
the two images is significant. Fig. 10(c) shows the image recon-
structed by the PWLS-PINL method using Fig. 10(a) as the prior
image. Fig. 10(d) shows the image reconstructed by the PWLS-
PINL method using Fig. 10(b) as the prior image. For both the
reconstructions of Fig. 10(c) and (d), h was 2.0 ×10−3 and β
was 2.0 × 104 . Fig. 10(e) shows the corresponding difference
image between Figs. 10(c) and 2(a). Fig. 10(f) shows the corre-
sponding difference image between Fig. 10(d) and 2(a). These
difference images indicate that the performance of the PWLS-
PINL method is not sensitive to the deformed prior image when
a large search window N is used to minimize the influence of
the misregistration. In other words, the present PWLS-PINL
Fig. 11. Influence of the mismatch between the first prior image and the
desired image on the performance of the PWLS-PINL and PICCS methods.
(a) the FBP image reconstructed with ramp filter from the low-dose sinogram
data; (b) unregistered prior image reconstructed from the normal dose sinogram
data using the FBP method with ramp filter; (c) image reconstructed by the
PICCS method from the low-dose sinogram data using (b) as the prior-image
(α = 0.7, λ = 8.0 × 10−3 ); and (d) image reconstructed by the PWLS-PINL
method from the low-dose sinogram data using (b) as the prior image (N = 35 ×
35, n = 5 × 5, h = 1.0 × 10−3 , β = 5.0 × 105 ). The zoomed ROIs indicated
by three squares in (c) and (d) are displayed at the bottom, respectively. All the
images are displayed in a same window.
Fig. 12. Influence of the mismatch between the second prior image and the
desired images on the PWLS-PINL and PICCS methods. (a) FBP image recon-
structed with ramp filter from the low-dose sinogram data; (b) unregistered prior
image reconstructed from the normal dose sinogram data using the FBP method
with ramp filter; (c) image reconstructed by the PICCS method from the low-
dose sinogram data using (b) as the prior image (α = 0.7, λ = 8.0 × 10−3 ); and
(d) image reconstructed by the PWLS-PINL method from the low-dose sinogram
data using (b) as the prior image (N = 35 × 35, n = 5 × 5, h = 1.0 × 10−3 ,
β = 5.0 × 105 ). The zoomed ROIs indicated by three squares in (c) and (d) are
displayed at the bottom, respectively. All the images are displayed in a same
window.
method does not heavily depend on the prior image registration
accuracy. In practice, a rough registration may be sufficient.
B. Numerical XCAT Phantom Study
The adjacent slices of the normal-dose images were used
as the prior image to induce the desired-image reconstruction
by the PICCS and PWLS-PINL methods from the low-dose
sinogram data. Figs. 11 and 12 show the resultant influence of
mismatch between the two different prior images and the desired
ZHANG et al.: ITERATIVE RECONSTRUCTION FOR X-RAY COMPUTED TOMOGRAPHY 2375

images on the PWLS-PINL and PICCS methods, respectively.

Figs. 11(a) and 12(a) show the FBP images reconstructed with
ramp filter from the low-dose sinogram data; Figs. 11(b) and
12(b) show the unregistered prior images reconstructed from
the normal-dose sinogram data by the FBP method with ramp
filter. The bone structure at the top of Fig. 11(a) does not ap-
pear in Fig. 11(b), but appears large in Fig. 12(b) due to the
different slice selections. These kinds of features were used to
analyze the influence of prior information on the desired-image
reconstruction. Figs. 11(c) and 12(c) show the images recon-
structed by the PICCS method from the low-dose sinogram data
using Figs. 11(b) and 12(b) as the prior image, respectively. The
zoomed ROIs at the bottom of each subfigure show the pseu-
doboundaries from the PICCS method. The results suggest that
some local mismatches between the prior and current images
may propagate into the desired image. However, Figs. 11(d)
and 12(d) show that the PWLS-PINL method with the same
prior image can deal with the mismatches between the prior
and current images. These results indicate that the performance
of the PWLS-PINL method is not sensitive to the mismatched
prior images when a relatively large search window is used to
minimize the influence of the mismatch regions. In other words,
the present PWLS-PINL method does not heavily depend on the
prior-image mismatch accuracy, and a rough registration may
be sufficient in practice.
For the low-contrast detectability, as shown in Fig. 11(c)-
III and 11(d)-III, the PICCS method appears to perform better
than the PWLS-PINL method in the matched regions. How-
Fig. 13. Clinical chest CT image reconstructions. (a) image reconstructed by
ever, as shown in the zoomed ROI of Figs. 11(c) and 12(c), the the FBP method with ramp filter from the sinogram data acquired at 200 mAs,
PWLS-PINL method outperforms the PICCS method when the 120 kVp, which is used as the prior image for the PWLS-PINL and PICCS
low contrast features in the desired-image are different from methods; (b) image reconstructed by the FBP with ramp filter from the sinogram
data acquired at 20 mAs, 120 kVp; (c) image reconstructed by the PWLS-NL
those in the prior image. method from the sinogram data acquired at 20 mAs, 120 kVp (N = 21 × 21,
n = 5 × 5, h = 0.6 × 10−3 , β = 1.0 × 104 ); (d) image reconstructed by the
PICCS method from the sinogram data acquired at 20 mAs, 120 kVp (α = 0.5,
C. Patient Study λ = 0.1); (e) image reconstructed by the GDSIR method from the sinogram data
acquired at 20 mAs, 120 kVp (γ = 4 × 10−2 ); and (f) image reconstructed by
Fig. 13 shows the chest CT images reconstructed by five dif- the PWLS-PINL method from the sinogram data acquired at 20 mAs, 120 kVp
ferent methods from two different sinogram data. Fig. 13(a) (N = 41 × 41, n = 5 × 5, h = 0.6 × 10−3 , β = 1.0 × 104 ). All the images
are displayed in the window [0 0.02] mm−1 .
shows the image reconstructed by the FBP method with ramp
filter from the sinogram data acquired at 200 mAs, 120 kVp,
which was used as the prior image for the PWLS-PINL, PICCS,
and GDSIR methods. Fig. 13(b) shows the image reconstructed
by the FBP with ramp filter from the sinogram data acquired at
20 mAs, 120 kVp. The anatomical structures between two FBP dictionary was learned from patches extracted from Fig. 13(a).
images are mostly similar except some noticeable organ elas- It can be seen that most of the noise is removed without introduc-
tic deformation as indicated by the rectangles in Fig. 13(a) and ing new fake structure, but noticeable streak artifacts still exist.
(b) because of the heteronomous breathing or heartbeat. The Fig. 13(f) shows the image reconstructed by the PWLS-PINL
noise-induced streak artifacts can be observed in Fig. 13(b). method from the sinogram data acquired at 20 mAs, 120 kVp. To
Fig. 13(c) shows the image reconstructed by the PWLS-NL illustrate further the performance of the PWLS-based methods,
method from the sinogram data acquired at 20 mAs, 120 kVp. the zoomed ROIs are shown in Fig. 14. The results show that
Fig. 13(d) shows the image reconstructed by the PICCS method the present PWLS-PINL method can achieve significant gains
using Fig. 13(a) as the prior image. Most of the noise is miti- over the other methods in terms of reconstruction accuracy by
gated by the TV term of the PICCS objective function, whereas visual inspection. In the implementation of the four PWLS-
some deceptive structures appear in the mismatched areas. This based methods, the FBP image reconstructed with ramp filter,
phenomenon suggests that the edge structure in the prior im- i.e., Fig. 13(b) was used as the initial image. All the parameters
ages may have been passed to the current images. Fig. 13(e) were selected by visual inspection for eye-appealing results, to-
shows the image reconstructed by the GDSIR method from gether with the normal-dose FBP image [see i.e., Fig. 13(a)] for
the sinogram data acquired at 20 mAs, 120 kVp. The global comparison.
2376 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 9, SEPTEMBER 2014
Fig. 14. Zoomed ROIs in Fig. 12. (a) image reconstructed by the PWLS-NL
method from the sinogram data acquired at 20 mAs, 120 kVp (N = 21 × 21,
n = 5 × 5, h = 0.6 × 10−3 , β = 1.0 × 104 ); (b) image reconstructed by the
PICCS method from the sinogram data acquired at 20 mAs, 120 kVp (α = 0.5,
λ = 0.1); (c) image reconstructed by the GDSIR method from the sinogram data
acquired at 20 mAs, 120 kVp (γ = 4 × 10−2 ); and (d) image reconstructed by
the PWLS-PINL method from the sinogram data acquired at 20 mAs, 120 kVp
(N = 41 × 41, n = 5 × 5, h = 0.6 × 10−3 , β = 1.0 × 104 ). All the images
are displayed in the window [0 0.02] mm−1 .
IV. DISCUSSION AND CONCLUSION
Prior information is useful for successful PWLS-based image
reconstruction [11], [22]–[24], [28]. In the medical clinic, sev-
eral situations exist where repeated tomographic acquisitions are
prescribed. In these cases, substantial similarity exists among
images in the acquisition sequence. As a result, a previously
scanned normal-dose CT image may provide strong prior infor-
mation for subsequent iterative image reconstruction from the
projection data with a low-dose protocol. For example, Chen
et al. [29] proposed the PICCS method for CT iterative image
reconstruction from the spare-view measurements using a high-
quality reference image. Ma et al. [10] proposed the ndiNLM
filter for low-dose CT image restoration using the previously
scanned CT image without requiring the accurate image reg-
istration processing due to its patch-based search mechanism.
In this paper, we have proposed a prior-image induced nonlo-
cal regularization term for PWLS image reconstruction, which
was named as the “PWLS-PINL.” Preliminary results using
both phantom and patient data suggested that the present al-
gorithm can robustly incorporate redundant information from
previously acquired images without requiring accurate image
registration, as discussed in Section III-A5. Moreover, a sub-
stantial improvement in image quality was observed compared
with the traditional methods if the prior images were not incor-
porated. The present algorithm can be applied to the scenario
where sequences of tomographic data are acquired, such as lung
nodule surveillance, image-guided biopsy needle, and volumet-
ric CBCT imaging during a radiation therapy course. Because
the prior image may be offset or rotated with respect to the
anatomy at the time of imaging, a coarse registration should
be applied in the implementation of the present PWLS-PINL
method aiming to reduce the computational burden with a se-
lected median size search window. Practically, due to the regis-
tration accuracy depending on the dose in CT scan [42], some
robust registration techniques should be adopted to improve the
registration accuracy with a low-dose CT imaging.
In this study, two prior image-based methods were explored
for comparison, i.e., the PICCS and GDSIR. The PICCS and
PWLS-PINL methods can be regarded as a voxel updating tech-
nique while the GDSIR method is a patch updating technique.
As shown in Section-III, the PICCS method cannot achieve sat-
isfactory performance in suppressing noise-induced artifacts as
the misalignments between the prior- and desired-images are
significant, and, the GDSIR method cannot also effectively deal
with the noise-induced artifacts in the case of ultra-low-dose
scan. Meanwhile, the present PWLS-PINL method can relax
the tradeoff of fake structures introduction and noise-induced
streak artifacts reduction based on the nature of nonlocal means
processing in image restoration. In addition, it is worth noting
that the nonlocal mean algorithm as a postprocessing technique
for low-dose CT image restoration can also work well in some
cases. But, when the projection data were seriously corrupted
by excessive X-ray photon noise, the associative FBP image
will suffer from serious noise-induced streak artifacts and then
the nonlocal means algorithm cannot yield any possible satisfac-
tory results due to its ability in preserving structural information.
Meanwhile, the present PWLS-PINL method, by modeling sev-
eral important physical and statistical factors including detector
geometry, photon statistics, and electronic background noise,
can yield more accurate results than the nonlocal means algo-
rithm from the noisy projection measurements.
For the implementation of the PWLS-PINL method, four
parameters, i.e., search window N , patch-window n, control
parameter h, and hyperparameter β were selected by visual
inspection for eye-appealing result, together with the normal-
dose image as comparison. In practice, to reduce the computing
burden, the size of the search window should be limited to an
appropriate size. However, the search window should be large
enough to obtain more similar information while minimizing
the influence of the mismatched tissues. Hence, search window
N should be optimally determined by fully considering both
the registration accuracy and the degree of the overall struc-
ture changes between the to-be-reconstructed and prior images.
For the selection of the patch-window, extensive experiments
showed that a 5 × 5 patch-window is adequate for effective
noise and artifacts suppression while retaining computational
efficiency. As for the selection of h and β in this study, we
briefly fixed the sizes of the search- and patch-windows and did
not consider the methods that adaptively optimized them, but
we studied the results obtained using a broad range of parame-
ter values by hand in term of visual inspection and quantitative
measurements. This scheme can also be considered as a trial
and error process. Theoretical studies in optimizing the nonlo-
cal relevant parameters have already been conducted using the
SURE approach [43], [44], which provided a useful technique
to determine the parameters in the present algorithm. This is an
interesting topic for future research.
For the modified iterative successive overrelaxation algorithm
in this study, we would like to make the following discussions.
As a version of the widely used one-step-late (OSL) iteration
scheme, the present PWLS-PINL algorithm can be feasibly and
effectively implemented using the present joint optimal recon-
struction strategy presented in Section II-C. However, due to the
fact that weight w(k, j) in (6) depends on the unknown image
intensity, the present PINL regularization in (5) is nonconvex for
ZHANG et al.: ITERATIVE RECONSTRUCTION FOR X-RAY COMPUTED TOMOGRAPHY
a global optimization. Similar to many existing OSL algorithms
where global convergence is an open issue, the present algo-
rithm also suffered from the lack of strict global convergence
proof. Despite the lack of guarantee of global convergence, the
present algorithm is still effective in practice together using the
joint optimization scheme.
Another major drawback of the present PWLS-PINL algo-
rithm is that the update of the nonlocal weights matrix unavoid-
ably increases the computational burden as described in [28].
To address this issue, one possible strategy is to use infrequent
update of the parameters (nonlocal weights matrix) interleaved
with more frequent OSL steps; and another one is to use fast
computers and dedicated hardware [45] to develop parallel it-
erative image reconstruction algorithm. We believe that in the
future, most iterative-based image reconstruction algorithms in-
cluding the present PWLS-PINL may be widely used in medical
clinic.
ACKNOWLEDGMENT
The authors would like to thank the anonymous reviewers
for their constructive comments and suggestions that greatly
improve the quality of the manuscript.
REFERENCES
[1] T. Schubert, A. L. Jacob, M. Takes, T. Menter, A. Gutzeit, and S. Kos,
“CT-guided percutaneous biopsy of a mass lesion in the upper presacral
space: A sacral transneuroforaminal approach,” Cardiovasc. Intervent.
Radiol., vol. 35, no. 5, pp. 1255–1257, Oct. 2012.
[2] J. H. Siewerdsen, “Cone-beam CT with a flat-panel detector: From image
science to image-guided surgery,” Nucl. Instrum. Methods Phys. Res. A,
vol. 648, no. S1, pp. S241–S250, Aug. 2011.
[3] S. B. Dharap, A. A. Khandkar, A. Pandey, and A. K. Sharma, “Repeat CT
scan in closed head injury,” Injury, vol. 36, no. 3, pp. 412–416, Mar. 2005.
[4] B. Zhao, L. P. James, C. S. Moskowitz, P. Guo, M. S. Ginsberg,
R. A. Lefkowitz, Y. Qin, G. J. Riely, M. G. Kris, and L. H. Schwartz,
“Evaluating variability in tumor measurements from same-day repeat CT
scans of patients with non-small cell lung cancer,” Radiology, vol. 252,
no. 1, pp. 263–272, Jul. 2009.
[5] Y. C. Kuo, T. H. Wu, T. S. Chung, K. W. Huang, K. S. Chao, W. C. Su,
and J. F. Chiou, “Effect of regression of enlarged neck lymph nodes on
radiation doses received by parotid glands during intensity-modulated
radiotherapy for head and neck cancer,” Amer. J. Clin. Oncol., vol. 29,
no. 6, pp. 600–605, Dec. 2006.
[6] L. Yu, X. Liu, S. Leng, J. M. Kofler, J. C. Ramirez-Giraldo, M. Qu,
J. Christner, J. G. Fletcher, and C. H. McCollough, “Radiation dose reduc-
tion in computed tomography: Techniques and future perspective,” Imag.
Med., vol. 1, no. 1, pp. 65–84, Oct. 2009.
[7] C. H. McCollough, M. R. Bruesewitz, and J. M. Kofler, Jr., “CT dose
reduction and dose management tools: Overview of available options,”
Radiographics, vol. 26, no. 2, pp. 503–512, Mar./Apr. 2006.
[8] M. K. Kalra, M. M. Maher, T. L. Toth, B. Schmidt, B. L. Westerman,
H. T. Morgan, and S. Saini, “Techniques and applications of automatic
tube current modulation for CT,” Radiology, vol. 233, no. 3, pp. 649–657,
Dec. 2004.
[9] L. Yu, M. R. Bruesewitz, K. B. Thomas, J. G. Fletcher, J. M. Kofler, and
C. H. McCollough, “Optimal tube potential for radiation dose reduction
in pediatric CT: Principles, clinical implementations, and pitfalls,” Radio-
graphics, vol. 31, no. 3, pp. 835–848, May/Jun. 2011.
[10] J. Ma, J. Huang, Q. Feng, H. Zhang, H. Lu, Z. Liang, and W. Chen, “Low-
dose computed tomography image restoration using previous normal-dose
scan,” Med. Phys., vol. 38, no. 10, pp. 5713–5731, Oct. 2011.
[11] J. Wang, T. Li, and L. Xing, “Iterative image reconstruction for CBCT
using edge-preserving prior,” Med. Phys., vol. 36, no. 1, pp. 252–260, Jan.
2009.
[12] J. Wang, T. Li, H. Lu, and Z. Liang, “Penalized weighted least-squares
approach to sinogram noise reduction and image reconstruction for low-
2377
dose X-ray computed tomography,” IEEE Trans. Med. Imag., vol. 25,
no. 10, pp. 1272–1283, Oct. 2006.
[13] S. Singh, M. K. Kalra, M. D. Gilman, J. Hsieh, H. H. Pien,
S. R. Digumarthy, and J. A. Shepard, “Adaptive statistical iterative re-
construction technique for radiation dose reduction in chest CT: A pilot
study,” Radiology, vol. 259, no. 2, pp. 565–573, May 2011.
[14] L. Ouyang, T. Solberg, and J. Wang, “Effects of the penalty on the penal-
ized weighted least-squares image reconstruction for low-dose CBCT,”
Phys. Med. Biol., vol. 56, no. 17, pp. 5535–5552, Sep. 2011.
[15] J. P. Guedon and Y. Bizais, “Bandlimited and Haar filtered back-projection
reconstructions,” IEEE Trans. Med. Imag., vol. 13, no. 3, pp. 430–440,
Sep. 1994.
[16] H. G. Hofmann, B. Keck, C. Rohkohl, and J. Hornegger, “Comparing
performance of many-core CPUs and GPUs for static and motion com-
pensated reconstruction of C-arm CT data,” Med. Phys., vol. 38, no. 1,
pp. 468–473, Jan. 2011.
[17] X. Jia, B. Dong, Y. Lou, and S. B. Jiang, “GPU-based iterative cone-beam
CT reconstruction using tight frame regularization,” Phys. Med. Biol.,
vol. 56, no. 13, pp. 3787–3807, Jul. 2011.
[18] L. M. Mitsumori, W. P. Shuman, J. M. Busey, O. Kolokythas, and
K. M. Koprowicz, “Adaptive statistical iterative reconstruction versus fil-
tered back projection in the same patient: 64 channel liver CT image quality
and patient radiation dose,” Eur. Radiol., vol. 22, no. 1, pp. 138–143, Jan.
2012.
[19] S. Z. Li, Markov Random Field Modeling in Image Analysis. Tokyo,
Japan: Springer-Verlag, 2001.
[20] D. Shulman and J Herve, “Regularization of discontinuous flow fields,”
in Proc. Workshop Vis. Motion, Irvine, CA, USA, 1989, pp. 81–86.
[21] M. Black and G. Sapiro, “Edges as outliers: Anisotropic smoothing using
local image statistics,” in Proc. Scale-Space Theories Comput. Vis., Corfu,
Greece, 1999, pp. 259–270.
[22] D. F. Yu and J. A. Fessler, “Edge-preserving tomographic reconstruction
with nonlocal regularization,” IEEE Trans. Med. Imag., vol. 21, no. 2,
pp. 159–173, Feb. 2002.
[23] J. Ma, Q. Feng, Y. Feng, J. Huang, and W. Chen, “Generalized Gibbs priors
based positron emission tomography reconstruction,” Comput. Biol. Med.,
vol. 40, no. 6, pp. 565–571, Jun. 2010.
[24] Y. Chen, D. Gao, C. Nie, L. Luo, W. Chen, X. Yin, and Y. Lin, “Bayesian
statistical reconstruction for low-dose X-ray computed tomography us-
ing an adaptive-weighting nonlocal prior,” Comput. Med. Imag. Graph.,
vol. 33, no. 7, pp. 495–500, Oct. 2009.
[25] H. Yu, S. Zhao, E. A. Hoffman, and G. Wang, “Ultra-low dose lung CT
perfusion regularized by a previous scan,” Acad. Radiol., vol. 16, no. 3,
pp. 363–373, Mar. 2009.
[26] A. Buades, B. Coll, and J. M. Morel, “A nonlocal algorithm for image
denoising,” in Proc. IEEE Comput. Vis. Pattern Recog, 2005, vol. 2,
pp. 60–65.
[27] J. Ma, Z. Liang, Y. Fan, Y. Liu, J. Huang, W. Chen, and H. Lu, “Vari-
ance analysis of x-ray CT sinograms in the presence of electronic noise
background,” Med. Phys., vol. 39, no. 7, pp. 4051–4065, Jul. 2012.
[28] J. Ma, H. Zhang, Y. Gao, J. Huang, Z. Liang, Q. Feng, and W. Chen,
“Iterative image reconstruction for cerebral perfusion CT using a pre-
contrast scan induced edge-preserving prior,” Phys. Med. Biol., vol. 57,
no. 22, pp. 7519–7542, Nov. 2012.
[29] G. H. Chen, P. Theriault-Lauzier, J. Tang, B. Nett, S. Leng, J. Zambelli,
Z. Qi, N. Bevins, A. Raval, S. Reeder, and H. Rowley, “Time-resolved
interventional cardiac C-arm cone-beam CT: An application of the PICCS
algorithm,” IEEE Trans. Med. Imag., vol. 31, no. 4, pp. 907–923, Apr.
2012.
[30] P. T. Lauzier and G. H. Chen, “Characterization of statistical prior image
constrained compressed sensing (PICCS)—Part II. Application to dose
reduction,” Med. Phys., vol. 40, no. 2, pp. 021902:1–021902:14, Feb.
2013.
[31] L. Ouyang, T. Solberg, and J. Wang, “Noise reduction in low-dose cone
beam CT by incorporating prior volumetric image information,” Med.
Phys., vol. 39, no. 5, pp. 2569–2577, May 2012.
[32] G. Han, Z. Liang, and J. You, “A fast ray-tracing technique for TCT and
ECT studies,” in Proc. IEEE Nucl. Sci. Symp., Seattle, WA, USA, 1999,
vol. 3, pp. 1515–1518.
[33] T. Li, X. Li, J. Wang, J. Wen, H. Lu, J. Hsieh, and Z. Liang, “Nonlinear
sinogram smoothing for low-dose X-ray CT,” IEEE Trans. Nucl. Sci.,
vol. 51, no. 5, pp. 2505–2513, Oct. 2004.
[34] Q. Xu, H. Yu, X. Mou, L. Zhang, J. Hsieh, and G. Wang, “Low-dose X-
ray CT reconstruction via dictionary learning,” IEEE Trans. Med. Imag.,
vol. 31, no. 9, pp. 1682–1697, Sep. 2012.
2378 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 9, SEPTEMBER 2014
[35] M. Elad and M. Aharon, “Image denoising via sparse and redundant
representations over learned dictionaries,” IEEE Trans. Image Process.,
vol. 15, no. 12, pp. 3736–3745, Dec. 2006.
[36] Y. C. Pati, R. Rezaiifar, and P. Krishnaprasad, “Orthogonal matching pur-
suit: Recursive function approximation with applications to wavelet de-
composition,” in Proc. 27th Asilomar Conf. Signals Syst. Comput., Pacific
Grove, CA, USA, 1993, vol. 1, pp. 40–44.
[37] R. Szeliski and J. Coughlan, “Spline-Based Image Registration,” Int. J.
Comput. Vis., vol. 22, no. 3, pp. 199–218, Mar./Apr. 1997.
[38] J. A. Fessler, “Penalized weighted least-squares image reconstruction for
positron emission tomography,” IEEE Trans. Med. Imag., vol. 13, no. 2,
pp. 290–300, Jun. 1994.
[39] W. P. Segars, G. Sturgeon, S. Mendonca, J. Grimes, and B. M. Tsui, “4D
XCAT phantom for multimodality imaging research,” Med. Phys., vol. 37,
no. 9, pp. 4902–4915, Sep. 2010.
[40] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality
assessment: From error visibility to structural similarity,” IEEE Trans.
Image Process., vol. 13, no. 4, pp. 600–612, Apr. 2004.
[41] M. J. Black, G. Sapiro, D. H. Marimont, and D. Heeger, “Robust
anisotropic diffusion,” IEEE Trans. Image Process., vol. 7, no. 3, pp. 421–
432, Mar. 1998.
[42] O. Dandekar, K. Siddiqui, V. Walimbe, and R. Shekhar, “Image registra-
tion accuracy with low-dose CT: How low can we go?” in Proc. IEEE Int.
Symp. Biomed. Imag., Arlington, VA, USA, 2006, pp. 502–505.
[43] D. V. De Ville and M. Kocher, “SURE-based non-local means,” IEEE
Signal Process. Lett., vol. 16, no. 11, pp. 973–976, Nov. 2009.
[44] V. Duval, J. F. Aujol, and Y. Gousseau, “On the parameter choice for the
non-local means,” Tech. Rep. HAL-00468856, 2010.
[45] F. Xu and K. Mueller, “Accelerating popular tomographic reconstruction
algorithms on commodity PC graphics hardware,” IEEE Trans. Nucl. Sci.,
vol. 52, no. 3, pp. 654–663, Jun. 2005.
Authors’ photographs and biographies not available at the time of publication.