Deep learning (DL) shows promise of advantages over conventional signal processing techniques in a variety of imaging applications. The networks’ being trained from examples of data rather than explicitly designed allows them to learn signal and noise characteristics to most effectively construct a mapping from corrupted data to higher quality representations. In inverse problems, one has options of applying DL in the domain of the originally captured data, in the transformed domain of the desired final representation, or both. X-ray computed tomography (CT), one of the most valuable tools in medical diagnostics, is already being improved by DL methods. Whether for removal of common quantum noise resulting from the Poisson-distributed photon counts, or for reduction of the ill effects of metal implants on image quality, researchers have begun employing DL widely in CT. The selection of training data is driven quite directly by the corruption on which the focus lies. However, the way in which differences between the target signal and measured data is penalized in training generally follows conventional, pointwise loss functions. This work introduces a creative technique for favoring reconstruction characteristics that are not well described by norms such as mean-squared or mean-absolute error. Particularly in a field such as X-ray CT, where radiologists’ subjective preferences in image characteristics are key to acceptance, it may be desirable to penalize differences in DL more creatively. This penalty may be applied in the data domain, here the CT sinogram, or in the reconstructed image. We design loss functions for both shaping and selectively preserving frequency content of the signal.
The iterative reconstruction methods ASiR and ASiR-V have been accepted by hundreds of sites as their standard of care for a variety of protocols and applications. While the reduction in noise has been significant some readers have a preference for the classic image appearance. To maintain the classic image appearance of FBP at the same dose levels used for the standard of care with ASiR-V we introduce, Deep Learning Image Reconstruction (DLIR), a technique using artificial neural networks. This paper demonstrates that DLIR can maintain or improve upon the performance of the conventional iterative reconstruction algorithm (ASiR-V) in terms of low contrast detectability, noise, and spatial resolution.
KEYWORDS: In vivo imaging, Numerical simulations, Data acquisition, Computed tomography, Fluctuations and noise, Infrared imaging, Statistical analysis, Image quality, Medical imaging, Zoom lenses
In recent years, iterative reconstruction methods have been investigated extensively with the aim of reducing radiation dose while maintaining image quality in CT exams. In such a case, redundant data is usually available. In conventional FBP-type reconstructions, redundant data has to be carefully treated by applying a redundant weighting factor, such as Parker weighting. However, such a redundant weight has not been fully studied in a statistical iterative reconstruction framework. In this work, both numerical simulations and in vivo data sets were analyzed to study the impact of redundant weighting schemes on the reconstructed images for both static and moving objects. Results demonstrated that, for a static object, there was no obvious difference in the iterative reconstructions using different redundant weighting schemes, because the redundant data was consistent, and therefore, they all converged to the same solution. On the contrary, for a moving object, due to the inconsistency of the data, different redundant weighting schemes converged to different solutions, depending on the weight given to the data. The redundant weighting, if appropriately selected, can reduce motion-induced artifacts.
Statistical Image Reconstruction (SIR) often involves a balance of two requirements: the first requirement is enforcing a minimal difference between the forward projection of the reconstructed image with the measured projection data and the second requirement enforcing some kind of image smoothness, which depends on the specific selection of regularizer, to reduce the noise in the reconstructed image. The needed delicate balance between these two requirements in the numerical implementations often slow down the reconstruction speed due to either a degradation in convergence rate of the algorithm or a degradation of parallellizability of the numerical implementation algorithms. In this work, a general numerical implementation strategy has been proposed to allow the SIR algorithms to be implemented in two decoupled and alternating steps. The first step using SIR without any regularizer which allows for the use of the well-known ordered subset (OS) strategy to accelerate the image reconstruction. The second step solves a denoising problem without involving the data fidelity term. The alternation of these two decoupled steps enable one to perform SIR with both high convergence rate and high parallellizability. The total variation norm of the image has been used as an example of regularizers to illustrate the proposed numerical implementation strategy. Numerical simulations have been performed to validate the proposed algorithm. The noise-spatial resolution tradeoff curve and convergence speed of the algorithm have been investigated and compared against the conventional gradient descent based implementation strategy.
The statistical model based iterative reconstruction (MBIR) method has been introduced to clinical CT systems. Due to the nonlinearity of this method, the noise characteristics of MBIR are expected to differ from those of filtered backprojection (FBP). This paper reports an experimental characterization of the noise performance of MBIR equipped on several state-of-the-art clinical CT scanners at our institution. The thoracic section of an anthropomorphic phantom was scanned 50 times to generate image ensembles for noise analysis. Noise power spectra (NPS) and noise standard deviation maps were assessed locally at different anatomical locations. It was found that MBIR lead to significant reduction in noise magnitude and improvement in noise spatial uniformity when compared with FBP. Meanwhile, MBIR shifted the NPS of the reconstructed CT images towards lower frequencies along both the axial and the z frequency axes. This effect was confirmed by a relaxed slice thicknesstradeoff relationship shown in our experimental data. The unique noise characteristics of MBIR imply that extra effort must be made to optimize CT scanning parameters for MBIR to maximize its potential clinical benefits.
KEYWORDS: Image restoration, Computed tomography, Image quality, CT reconstruction, Reconstruction algorithms, Compressed sensing, Temporal resolution, Medical imaging, Data acquisition, Signal to noise ratio
Iterative image reconstruction methods have been proposed in computed tomography to address two major
challenges: one is to reduce radiation dose while maintaining image quality and the other is to reconstruct
diagnostic quality images from angularly sparse projection datasets. A variety of regularization models have
been introduced in these iterative image reconstruction methods to incorporate the desired image features. To
address the sparse view angle image reconstruction problem in four-dimensional cone-beam CT (4DCBCT),
Prior Image Constrained Compressed Sensing (PICCS) was proposed. In the past in 4DCBCT, as well as
other applications of the PICCS algorithm, the PICCS regularization was formulated using the 1 norm as
the means to promote image sparsity. The 1 norm in the objective function is not differentiable and thus
may pose challenges in numerical implementations. When the norm deviates from 1.0, the differentiability of
the objective function improves, however, the imaging performance may degrade in image reconstruction from
sparse datasets. In this paper, we study how the performance of PICCS-4DCBCT changes with norm selection
and whether the introduction of a reweighted scheme in relaxed norm PICCS reconstruction helps improve the
imaging performance.
Dual-energy CT has the potential to overcome many of the limitations of routine single-energy CT scanning, such a,.., the potential to provide quantitative imaging via electron density, effective atomic munber, and virtual monochromatic imaging and the potential to completely eliminate beam-hardening artifacts via projection space decomposition. While the potential clinical benefit is strong, a possible barrier to more frequent clinical use of dual-energy CT scanning is radiation dose for high quality images. While image quality in dual-energy CT depends on a munber of factors, including dose partitioning, the choice of kV pair, and the amount of pre filtration used, a munber of strategies have been employed to improve image quality in dual-energy CT. Four main methods are: (1) increa,..,e the radiation dose, (2) increase the slice thickness, (3) perform voxel averaging, or (4) use noise reduction algorithms. While these methods offer options for improving image quality, ideally, it is desirable not to have to increase radiation dose or sacrifice spatial resolution (in the x-y plane or in the z-direction). Therefore, it is the purpose of this work to investigate the application of Prior Image Constrained Compressed Sensing (PICCS) in dual-energy CT to reduce radiation dose without sacrificing image quality. In particular, we investigate the use of PICCS in dual-energy CT to generate material density images at half the radiation dose of a commonly used gemstone spectral imaging (GSI) protocol. lVIaterial density images are generated using half the radiation dose, and virtual monochromatic images are generated as a linear combination of half-dose material density images. In this abstract, qualitative and quantitative evaluation are provided to assess the performance of PICCS relative to FBP images at the full dose level and at the half dose level.
In this work we applied dose reduction using the prior image constrained compress sensing (DR-PICCS) method
on a C-arm cone beam CT system. DR-PICCS uses a smoothed image as the prior image. After applying DRPICCS,
the final image will have noise variance inherited from the prior image and spatial resolution from the
projection data. In order to investigate the dose reduction of DR-PICCS, three different dose levels were used in
C-arm scans of animal subjects using a Siemens Zeego C-arm system under an IACUC protocol. Image volumes
were reconstructed using the standard FBP and DR-PICCS algorithms(total of 160 images). These images were
randomly mixed and presented to three experienced interventional radiologists(each having more than twenty
years reading experience) to review and score using a five-point scale. After statistical significance testing, the
results show that DR-PICCS can achieve more than 60% dose reduction while keeping the same image quality.
And if we compare FBP and DR-PICCS at the same dose level the results show that DR-PICCS will generate
higher quality images.
Dose reduction using prior image constrained compressed sensing (DR-PICCS) is a method of CT reconstruction which utilizes prior image information in a compressed sensing framework to significantly reduce the noise in images acquired at low dose. The purpose of this study was to investigate the impact of edge sharpness and noise level in the prior image data on the resulting DR-PICCS images. Projection data from a 100 rnA CT myocardial perfusion scan of a swine was combined with numerically simulated projections of vessels of varying geometry (diameter = 4, 3, 2 mm) and contrast levels (600, 400, 200 HU enhancement). Bilateral and mean filters were applied to generate prior images, which were then used with the DR-PICCS algorithm. Vessel diameter, effective blurring kernel, and vessel intensity were compared among prior images as well as among the corresponding PICCS images. Although the filters produced prior images with significantly different spatial resolution characteristics at similar noise levels, these differences were mitigated in DR-PICCS images and the DR-PICCS had improved fidelity in comparison to the priors.
It is highly desirable to obtain perfusion information with the C-arm CBCT system in the interventional room. However,
due to hardware limitations, it is still elusive to achieve cone-beam CT perfusion measurements. In this study, we
performed a systematic study to investigate what the main limiting factors are that need to be addressed for future C-arm
cone beam CT perfusion imaging. To do so, we performed systematic numerical simulation studies using a diagnostic
CT perfusion data set. Specifically, a forward projection was performed to simulate cone-beam CT perfusion experiment
with C-arm CBCT geometry and temporal behavior. Different x-ray delays after contrast injection have been simulated
with this method. The view angle undersampling artifacts, shading artifacts from dynamic objects, and the importance of
arterial input function (AIF) for perfusion study were investigated in this study with different x-ray delay times. From the
simulation results, it was found that the view angle undersampling artifacts do not have much impact on perfusion maps.
The shading artifacts from dynamic object were shown to have a negligible effect on the NRMSE in perfusion maps. The
accuracy of AIF is an important but not a dominating factor for perfusion studies. C-arm CBCT cannot accurately
recover the slowly changing contrast in brain tissues due to the low temporal resolution. Therefore, to enable cone beam
C-arm CT perfusion measurement, it is critical to improve the temporal behavior of CBCT by either employing new
hardware upgrades or introducing new software methods.
In the current workflow of ischemic stroke management, it is highly desirable to obtain perfusion information with the
C-arm CBCT system in the interventional room. Due to hardware limitations, the data acquisition speed of the current Carm
CBCT systems is relatively slow and only 7 time frames are available for a 45 s perfusion study. In this study, a
novel temporal recovery method was proposed to recover contrast enhancement curves in C-arm CBCT perfusion
studies. The proposed temporal recovery problem is a constrained optimization problem. Two numerical methods were
used to solve the proposed problem. The feasibility of proposed temporal recovery method was validated with numerical
experiments. Both solvers can achieve a satisfactory solution for the temporal recovery problem, while the result of the
Bregman algorithm is more accurate than that from the CG. In vivo animal studies were used to demonstrated the
improvement of the proposed method in C-arm CBCT perfusion. A stoked canine model was scanned with both C-arm
CBCT and diagnostic CT. Perfusion defects can be clearly indentified from the cerebral blood flow (CBF) map of
diagnostic CT perfusion. Without the temporal recovery technique, these defects can hardly be identified from the CBCT CBF map. After applying the proposed temporal recovery method, the CBCT CBF map well correlates with the CBF
map from diagnostic CT.
Recently, the Statistical Image Reconstruction (SIR) and compressed sensing (CS) framework has shown promise
in the x-ray computed tomography (CT) community. In this paper, we propose to establish an equivalence
between the unconstrained optimization problem and a constrained optimization with explicit data consistency
term. The immediate consequence of the equivalence is to enable one to use the well-developed optimization
method to solve the constrained optimization problem to refine the solution of the corresponding unconstrained
optimization problem. As an application of this equivalence, the method was used to develop a convergent and
numerically efficient implementation for the prior image constrained compressed sensing (PICCS).
Advances have been made in recent years in computed tomography (CT) as a result of the development and
implementation of new iterative reconstruction methods. Prior Image Constrained Compressed Sensing (PICCS)
is one such iterative reconstruction method which iteratively minimizes an objective function to approach a target
image. To date, published studies have employed the L1 norm in the minimization of the objective function. In
this study, we investigate the use of Lp norms with p > 1 and investigate how image quality depends on the selection of the Lp norm used in the minimization of the objective function.
In recent years, iterative image reconstruction algorithms have received much interest in x-ray CT imaging. Images
reconstructed by the conventional filtered backprojection algorithm are often used as seed images to start the iterations.
This paper presents a new consistency property of the measured CT projection data and the forward projected data from
a motion contaminated image reconstructed using an analytical image reconstruction algorithm. It is theoretically proven
and numerically validated that, when the measured projection data is not redundant, the measured projection data is
consistent with the forward projected data of a reconstructed image using analytical image reconstruction algorithms, no
matter whether motion artifacts are present in the image or not. However, when there is redundancy in the measured
projection data, the consistency depends on the choice of weighting function used for the redundant data. The forward
projected data is always consistent with those measured projections with the weight 1.0, no matter what weighting
schemes are used. For the measured projection data with a weight smaller than 1.0, the forward projected data is
consistent with the weighted sum of redundantly measured projection data. With this new property, some potential issues
of directly using FBP images as seed images for iterative algorithms are discussed.
KEYWORDS: In vivo imaging, Image restoration, Computed tomography, Image filtering, Reconstruction algorithms, Tissues, Blood, Signal attenuation, Phase modulation, Analytical research
Myocardial perfusion scans are an important tool in the assessment of myocardial viability following an infarction.
Cardiac perfusion analysis using CT datasets is limited by the presence of so-called partial scan artifacts. These
artifacts are due to variations in beam hardening and scatter between different short-scan angular ranges. In this
research, another angular range dependent effect is investigated: non-uniform noise spatial distribution. Images
reconstructed using filtered backprojection (FBP) are subject to this effect. Statistical image reconstruction
(SIR) is proposed as a potential solution. A numerical phantom with added Poisson noise was simulated and
two swines were scanned in vivo to study the effect of FBP and SIR on the spatial uniformity of the noise
distribution. It was demonstrated that images reconstructed using FBP often show variations in noise on the
order of 50% between different time frames. This variation is mitigated to about 10% using SIR. The noise level
is also reduced by a factor of 2 in SIR images. Finally, it is demonstrated that the measurement of quantitative
perfusion metrics are generally more accurate when SIR is used instead of FBP.
The appeal of compressed sensing (CS) in the context of medical imaging is undeniable. In MRI, it could
enable shorter acquisition times while in CT, it has the potential to reduce the ionizing radiation dose imparted
to patients. However, images reconstructed using a CS-based approach often show an unusual texture and a
potential loss in spatial resolution. The prior image constrained compressed sensing (PICCS) algorithm has been
shown to enable accurate image reconstruction at lower levels of sampling. This study systematically evaluates
an implementation of PICCS applied to myocardial perfusion imaging with respect to two parameters of its
objective function. The prior image parameter α was shown here to yield an optimal image quality in the range
0.4 to 0.5. A quantitative evaluation in terms of temporal resolution, spatial resolution, noise level, noise texture,
and reconstruction accuracy was performed.
Radiation dose reduction remains at the forefront of research in computed tomography. X-ray tube parameters such as
tube current can be lowered to reduce dose; however, images become prohibitively noisy when the tube current is too
low. Wavelet denoising is one of many noise reduction techniques. However, traditional wavelet techniques have the
tendency to create an artificial noise texture, due to the nonuniform denoising across the image, which is undesirable
from a diagnostic perspective. This work presents a new implementation of wavelet denoising that is able to achieve
noise reduction, while still preserving spatial resolution. Further, the proposed method has the potential to improve those
unnatural noise textures. The technique was tested on both phantom and animal datasets (Catphan phantom and timeresolved
swine heart scan) acquired on a GE Discovery VCT scanner. A number of tube currents were used to
investigate the potential for dose reduction.
A technique for dose reduction using prior image constrained compressed sensing (DR-PICCS) in computed
tomography (CT) is proposed in this work. In DR-PICCS, a standard FBP reconstructed image is forward
projected to get a fully sampled projection data set. Meanwhile, it is low-pass filtered and used as the prior
image in the PICCS reconstruction framework. Next, the prior image and the forward projection data are
used together by the PICCS algorithm to obtain a low noise DR-PICCS reconstruction, which maintains the
spatial resolution of the original FBP images. The spatial resolution of DR-PICCS was studied using a Catphan
phantom by MTF measurement. The noise reduction factor, CT number change and noise texture were studied
using human subject data consisting of 20 CT colonography exams performed under an IRB-approved protocol.
In each human subject study, six ROIs (two soft tissue, two colonic air columns, and two subcutaneous fat)
were selected for the CT number and noise measurements study. Skewness and kurtosis were used as figures of
merit to indicate the noise texture. A Bland-Altman analysis was performed to study the accuracy of the CT
number. The results showed that, compared with FBP reconstructions, the MTF curve shows very little change
in DR-PICCS reconstructions, spatial resolution loss is less than 0.1 lp/cm, and the noise standard deviation
can be reduced by a factor of 3 with DR-PICCS. The CT numbers in FBP and DR-PICCS reconstructions agree
well, which indicates that DR-PICCS does not change CT numbers. The noise textures indicators measured
from DR-PICCS images are in a similar range as FBP images.
Recently, iterative image reconstruction algorithms have been extensively studied in x-ray CT in order to produce
images with lower noise variance and high spatial resolution. However, the images thus reconstructed often
have unnatural image noise textures, the potential impact of which on diagnostic accuracy is still unknown.
This is particularly pronounced in total-variation-minimization-based image reconstruction, where the noise
background often manifests itself as patchy artifacts. In this paper, a quantitative noise texture evaluation
metric is introduced to evaluate the deviation of the noise histogram from that of images reconstructed using
filtered backprojection. The proposed texture similarity metric is tested using TV-based compressive sampling
algorithm (CSTV). It was demonstrated that the metric is sensitive to changes in the noise histogram independent
of changes in noise level. The results demonstrate the existence tradeoff between the texture similarity metric
and the noise level for the CSTV algorithm, which suggests a potential optimal amount of regularization. The
same noise texture quantification method can also be utilized to evaluate the performance of other iterative
image reconstruction algorithms.
The Prior Image Constrained Compressed Sensing (PICCS) algorithm (Med. Phys. 35, pg. 660, 2008)
has been applied to several computed tomography applications with both standard CT systems and
flat-panel based systems designed for guiding interventional procedures and radiation therapy treatment
delivery. The PICCS algorithm typically utilizes a prior image which is reconstructed via the standard
Filtered Backprojection (FBP) reconstruction algorithm. The algorithm then iteratively solves for the
image volume that matches the measured data, while simultaneously assuring the image is similar to the
prior image. The PICCS algorithm has demonstrated utility in several applications including: improved
temporal resolution reconstruction, 4D respiratory phase specific reconstructions for radiation therapy,
and cardiac reconstruction from data acquired on an interventional C-arm. One disadvantage of the PICCS algorithm, just as other iterative algorithms, is the long computation times typically associated with reconstruction. In order for an algorithm to gain clinical acceptance reconstruction must be achievable in minutes rather than hours. In this work the PICCS algorithm has been implemented on the GPU in order to significantly reduce the reconstruction time of the PICCS algorithm. The Compute Unified Device Architecture (CUDA) was used in this implementation.
KEYWORDS: Reconstruction algorithms, Sensors, Computed tomography, Image restoration, Signal attenuation, Scanners, In vivo imaging, X-rays, Medical imaging, Data acquisition
C-arm CT is used in neurovascular interventions where a large flat panel detector is used to acquire cone-beam projection data. In this case, data truncation problems due to the limited detector size are mild. When the cone beam CT method is applied to cardiac interventions severe data truncation artifacts reduce the clinical
utility of the reconstructions. However, accurate reconstruction is still possible given a priori knowledge of the reconstruction values within a small region inside the FOV. Several groups have studied the case of the interior problem where data is truncated from all views. In this paper, we applied these new mathematical discoveries
to C-arm cardiac cone-beam CT to demonstrate that accurate image reconstruction may be achieved for cardiac
interventions. The method is applied to iteratively reconstruct the image volume such that it satisfies several
physical conditions. In this work, the algorithm is applied to data from in-vivo cardiac canine studies collected
using a clinical C-arm system. It is demonstrated that the algorithm converges well to the reconstruction values
of non-truncated data reconstructed using the FDK algorithm. Furthermore, proper convergence is achieved
by using only an estimate of the average value within a subregion as a priori information (i.e. the exact value
at each pixel in the a priori region need not be known). Two methods for obtaining a priori information are
compared.
A technique for temporal resolution improvement using prior image constrained compressed sensing (TRI-PICCS) in
multi-detector computed tomography (MDCT) cardiac imaging is proposed. In this work, the performance of TRIPICCS
was studied using a hybrid phantom which consists of realistic cardiac anatomy and objects moving with
designed trajectories. Several simulated moving vessels were added to different locations in the heart. Different motion
directions and simulated heart rates were investigated using half of the projection data of the short-scan angular range in
TRI-PICCS. Different angular ranges of projection data were also investigated in TRI-PICCS to evaluate the highest achievable temporal resolution. The results showed that the temporal improvement of TRI-PICCS is independent of the locations of the moving objects and motion directions. The motion artifacts at 100 bmp simulated heart rate can be significantly improved using TRI-PICCS compared with conventional filtered back projection (FBP). The minimum angular range requirement of TRI-PICCS is about 90°, corresponding to a temporal resolution improvement factor of 2.6
compared with the standard short-scan FBP reconstruction.
Of all available reconstruction methods, statistical iterative reconstruction algorithms appear particularly promising since
they provide accurate physical noise modeling. The newly developed compressed sensing (CS) algorithm has shown the
potential to accurately reconstruct images from highly undersampled data. In x-ray CT reconstructions, the CS algorithm
can be implemented in the statistical reconstruction framework. In this study, we compared the performance of two
standard statistical reconstruction algorithms (penalized weighted least square and q-GGMRF) to the CS algorithm. In
assessing the image quality using these non-linear reconstructions it is critical to utilize realistic background anatomy as
the reconstruction results are object dependent. A cadaver head was scanned on a Varian Trilogy system at different
dose levels. A quality factor which accounts for the noise performance and the spatial resolution was introduced to
objectively evaluate the performance of the algorithm under two conditions: 1) constant undersampling factor comparing
different algorithms at different dose levels and 2) varying undersampling factors and dose levels for the CS algorithm.
To facilitate this comparison the original CS method was also formulated in the framework of the statistical image
reconstruction algorithm. This is also a novel aspect of this work. Important conclusions of the measurements are that:
for realistic anatomy over 100 projections are needed to avoid streak artifacts even with CS reconstruction, regardless of
the algorithm employed it is beneficial to distribute the total dose to many views as long as each view remains quantum
noise limited, and the CS method is not appropriate for low dose levels because while it can mitigate streaking artifacts
the images being to exhibit a patchy behavior.
C-arm based cone-beam CT (CBCT) has evolved into a routine clinical imaging modality to provide threedimensional
tomographic image guidance before, during, and after an interventional procedure. It is often used
to update the clinician to the state of the patient anatomy and interventional tool placement. Due to the
repeatedly use of CBCT, the accumulated radiation dose in an interventional procedure has become a concern.
There is a strong desire from both patients and health care providers to reduce the radiation exposure required
for these exams. The overall objective of this work is to propose and validate a method to significantly reduce
the total radiation dose used during a CBCT image guided intervention. The basic concept is that the first
cone-beam CT scan acquired at the full dose will be used to constrain the reconstruction of the later CBCT
scans acquired at a much lower radiation dose. A recently developed new image reconstruction algorithm, Prior
Image Constrained Compressed Sensing (PICCS), was used to reconstruct subsequent CBCT images with lower
dose. This application differs from other applications of the PICCS algorithm, such as time-resolved CT or fourdimensional
CBCT (4DCBCT), because the patient position may be frequently changed from one CBCT scan
to another during the procedure. Thus, an image registration step to account for the change in patient position
is indispensable for use of the PICCS image reconstruction algorithm. In this paper, the image registration step
is combined with the PICCS algorithm to enable radiation dose reduction in CBCT image guided interventions.
Experimental results acquired from a clinical C-arm system using a human cadaver were used to validate the
PICCS algorithm based radiation dose reduction scheme. Using the proposed method in this paper, it has
been demonstrated that, instead of 300 view angles, this technique requires about 20 cone-beam view angles to
reconstruct CBCT angiograms. This signals a radiation dose reduction by a factor of approximately fifteen for
subsequent acquisitions.
Purpose: To achieve three dimensional isotropic dynamic cardiac CT imaging with high temporal resolution for
evaluation of cardiac function with a slowly rotating C-arm system.
Method and Materials: A recently introduced extension to compressed sensing, viz. Prior Image Constrained
Compressed Sensing (PICCS), in which a prior image is used as a constraint in the reconstruction has enabled this
application. An in-vivo animal experiment (e.g. a beagle model) was conducted using an interventional C-arm system.
The imaging protocol was as follows: contrast was injected, the contrast equilibrated, breathing was suspended for ~14
seconds during which time 420 equally spaced projections were acquired. This data set was used to reconstruct a fully
sampled blurred image volume using the conventional FDK algorithm (e.g. the prior image). Then the data set was
retrospectively gated into 19 phases according to the recorded ECG signal (heart rate ~ 95bpm) and images were
reconstructed with the PICCS algorithm.
Results: Cardiac MR was used as the gold standard due to its high temporal resolution. The same short-axis slice was
selected from the PICCS-CT data set and the MR data set. Manual contouring on the peak systolic and peak diastolic
frames was performed to assess the ejection fraction contribution from this single plane. The calculated ejection
fractions with PICCS-CT agreed well with the MR results.
Conclusion: We have demonstrated the ability to use a slowly rotating interventional C-arm system in order to make
measurements of cardiac function. The new technique provides high isotropic spatial resolution (~0.5 mm) along with
high temporal resolution (~ 33 ms). The evaluation of cardiac function demonstrated a great agreement with single slice
cardiac MR.
Recently, foundational mathematical theory, compressed sensing (CS), has been developed which enables accurate
reconstruction from greatly undersampled frequency information (Candes et. al. and Donoho). Using numerical
phantoms it has been demonstrated that CS reconstruction (e.g. minimizing the ℓ1 norm of the discrete gradient
of the image) offers promise for computed tomography. However, when using experimental CT projection data the
undersampling factors enabled were smaller than in numerical simulations. An extension to CS has recently been
proposed wherein a prior image is utilized as a constraint in the image reconstruction procedure (i.e. Prior Image
Constrained Compressed Sensing - PICCS). Experimental results are demonstrated here from a clinical C-arm
system, highlighting one application of PICCS in reducing radiation exposure during interventional procedures
while preserving high image quality. In this study a range of view angles has been investigated from very limited
angle aquisitions (e.g. tomosythesis) to undersampled CT acquisitions.
KEYWORDS: Reconstruction algorithms, Image restoration, Data acquisition, Image quality, Compressed sensing, Heart, Medical imaging, Signal to noise ratio, Magnetic resonance imaging, Numerical simulations
It has been known for a long time that, in order to reconstruct a streak-free image in tomography, the sampling of view
angles should satisfy the Shannon/Nyquist criterion. When the number of view angles is less than the Shannon/Nyquist
limit, view aliasing artifacts appear in the reconstructed images. Most recently, it was demonstrated that it is possible to
accurately reconstruct a sparse image using highly undersampled projections provided that the samples are distributed at
random. The image reconstruction is carried out via an l1 norm minimization procedure. This new method is generally referred to as compressed sensing (CS) in literature. Specifically, for an N×N image with significant image pixels, the number of samples for an accurate reconstruction of the image is . In medical imaging, some prior images may be reconstructed from a different scan or from the same acquired
time-resolved data set. In this case, a new
image reconstruction method, Prior Image Constrained Compressed Sensing (PICCS), has been recently developed to
reconstruct images using a vastly undersampled data set. In this paper, we introduce the PICCS algorithm and
demonstrate how to use this new algorithm to solve problems in medical imaging.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.