Misalignment-Correction in C-arm-based flat-detector CT (FD-CT) is a frequently discussed problem. To avoid artifacts
caused by geometrical instabilities, numerous methods for misalignment correction were investigated. Most of them
make use of a foregoing calibration routine, based on scanning a specific phantom. The aim of this study is to develop
and evaluate an online image-content-based calibration technique without using any kind of marker or calibration
phantom. The introduced method is based on a gradient descent method, minimizing an entropy criterion which is used
to optimize the underlying geometry parameters of the acquisition system. It is formed as multistep approach, including a
global, local and projection wise optimization. This enables the elimination of general system misalignments, as well as a
reduction of streak artifacts and the adjustment of patient motion artifacts. Phantom and patient measurements with the
C-arm FD-CT system Artis Zeego (Siemens AG, Healthcare Sector, Forchheim, Germany) were used to validate the
algorithm for realistic applications. It reduced most of the actual misalignment and increased image quality drastically.
Phantom-studies, starting from the standard system geometry without a foregoing calibration showed very good results.
Online-calibration is possible with our approach and therefore, the limitation to predefined scan-protocols is obsolete.
The evaluation of patient datasets brought out the same conclusions and provides the implication of simultaneous patient
Ring artifacts often appear in flat-detector CT because of imperfect or defect detector elements or calibration.
In high-spatial resolution CT images reducing such artifacts becomes a necessity. In this paper, we used the
post-processing ring correction in polar coordinates (RCP)1 to eliminate the ring artifacts. The median filter is
applied to the uncorrected images in polar coordinates and ring artifacts are extracted from the original images.
The algorithm has a very high computational cost due to the time-expensive median filtering and coordinate
transformation on CPUs. Graphics processing units (GPUs)ca n be seen as parallel co-processors with high
computational power. All steps of the RCP algorithm were implemented with CUDA2(Compute Unified Device
Architecture, NVIDIA). We introduced a new GPU-based branchless vectorized median (BVM)filter. 3, 4 This
algorithm is based on minmax sorting and keeps track of a sorted array from which values are deleted and to
which new values are inserted. For comparison purpose a modified pivot median filter5 on GPUs was presented,
which compares a pivot element to all other values and recursively finds the median element. We evaluated the
performance of the RCP method using 512 slices, each slice consisted of 512×512 pixels. This post-processing
method efficiently reduces ring artifacts in the reconstructed images and improves image quality. Our CUDAbased
RCP is up to 13.6 times faster than the optimized CPU-based (single core)r outine. Comparing our two
GPU-based median filters showed a performance benefit by roughly 60% when switching from Pivot to BVM
code. The main reason is that the BVM algorithm is branchless and makes use of data-level parallelism. The
BVM method is better suited to the model of modern graphics processing. A multi-GPU solution showed that
the performance scaled nearly linearly.
The high flexibility of C-arm flat-detector computed tomography (FDCT) is used in a volume of interest (VOI) imaging
method to handle the challenges of increasing spatial resolution, reducing noise and saving dose.
A low-dose overview scan of the object and a high-dose scan of an arbitrary VOI are combined. The first scan is
adequate for orientation to select the VOI and the second scan assures high image quality in the VOI. The combination is
based on a forward projection of the reconstructed overview volume and the measured VOI data in the raw data domain.
Differences in the projection values are matched before a standard Feldkamp-type reconstruction is performed. In
simulations, spatial resolution, noise and contrast detectability were evaluated. Measurements of an anthropomorphic
phantom were used to validate the proposed method for realistic application. In Monte Carlo dose simulations the dose
reduction potential was investigated.
By combination of the two scans an image is generated which covers the whole object and provides the actual VOI at
high image quality. Spatial resolution was increased whereas noise was decreased from outside to inside the VOI, e.g. for
the simulations from 0.8 lp/mm to 3.0 lp/mm and from 39 HU to 18 HU, respectively. Simultaneously, the cumulative
dose for this two-scan procedure was significantly reduced in comparison to a conventional high dose scan, e.g. for the
performed simulations and measurements by about 95 %.
The proposed VOI approach offers significant benefits with respect to high-resolution and low-contrast imaging of a
VOI at reduced dose.
Typical cupping correction methods are pre-processing methods which require either pre-calibration measurements
or simulations of standard objects to approximate and correct for beam hardening and scatter. Some
of them require the knowledge of spectra, detector characteristics, etc. The aim of this work was to develop a
practical histogram-driven cupping correction (HDCC) method to post-process the reconstructed images. We
use a polynomial representation of the raw-data generated by forward projection of the reconstructed images;
forward and backprojection are performed on graphics processing units (GPU). The coefficients of the polynomial
are optimized using a simplex minimization of the joint entropy of the CT image and its gradient. The
algorithm was evaluated using simulations and measurements of homogeneous and inhomogeneous phantoms.
For the measurements a C-arm flat-detector CT (FD-CT) system with a 30×40 cm2 detector, a kilovoltage on
board imager (radiation therapy simulator) and a micro-CT system were used. The algorithm reduced cupping artifacts both in simulations and measurements using a fourth-order polynomial and was in good agreement to the reference. The minimization algorithm required less than 70 iterations to adjust the coefficients only performing a linear combination of basis images, thus executing without time consuming operations. HDCC reduced cupping artifacts without the necessity of pre-calibration or other scan information enabling a retrospective improvement of CT image homogeneity. However, the method can work with other cupping correction algorithms or in a calibration manner, as well.
Metallic implants are responsible for various artifacts in
flat-detector computed tomography visible as streaks
and dark areas in the reconstructed volumetric images. In this paper a novel method for a fast reduction
of these metal artifacts is presented using a three-step correction procedure to approximate the missing parts
of the raw data. In addition to image quality aspects, this paper deals with the problem of high correction
latencies by proposing a reconstruction and correction framework, that utilizes the massive computational power
of graphics processing units (GPUs). An initial volume is reconstructed, followed by a 3-dimensional metal voxel
segmentation algorithm. These metal voxels allow us to identify
metal-influenced detector elements by using a
simplified geometric forward projection. Consequently, these areas are corrected using a 3D interpolation scheme
in the raw data domain, followed by a second reconstruction. This volume is then segmented into three materials
with respect to bone structures using a threshold-based algorithm. A forward projection of the obtained tissueclass
model substitutes missing or corrupted attenuation values for each detector element affected by metal and
is followed by a final reconstruction. The entire process including the initial reconstruction, takes less than a
minute (5123 volume with 496 projections of size 1240x960) and offers significant improvements of image quality.
The method was evaluated with data from two FD-CT C-arm systems (Artis Zee and Artis Zeego, Siemens
Healthcare, Forchheim, Germany).
To perform a perspective cone-beam backprojection of the
Feldkamp-type (FDK) the geometry of the approximately circular scan trajectory has to be available. If the system or the scan geometry is unknown and afflicted with geometric instabilities (misalignment) reconstructing a misaligned scan can cause severe artifacts in the
CT images. We propose an online and image-based iterative correction of a misaligned reconstruction geometry by using entropy minimization. Unlike current methods which use a calibration of the geometry for defined scan protocols and calibration phantoms, the proposed method is performed combining a simplex algorithm
for multi-parameter optimization and a graphics card (GPU)-based
FDK-reconstruction in an iterative scheme.
The simplex algorithm changes the geometric parameters of source and detector with respect to the reduction of entropy. In order to reduce the size of the dimensional space required for minimization the trajectory described by a subset of trajectory points. A virtual trajectory of an approximately circular path is generated after each
iteration of the algorithm. This method was validated using simulations and measurements performed on a Carm CT System equipped with a flat-panel detector (Axiom Artis, Siemens Healthcare, Forchheim, Germany).
Entropy was minimal for the ideal dataset, whereas strong misalignment resulted in a higher entropy value. The use of the
GPU-based reconstruction provided an online geometry correction after a total computation time of only 1-3 s using 100 to 300 iterations of the algorithm, depending on the degree of misalignment and initialization conditions.
We offer a novel approach for real-time CT reconstruction with the possibility of interactively changing parameters
like the position and orientation of the slice to arbitrary values by the user during the analysis. To
achieve this, a new reconstruction, including backprojection, is done every time the user wants to see a different
view (in contrast to computing a volume upfront). The reconstruction was implemented on a GPU (graphics
processing unit) using OpenGL and provides near real-time performance with less than 20 ms reconstruction
time for 512 × 512 images. With this approach the user is free to change parameters that are fixed when a
conventional reconstruction is used. So he is free to set the position of the slice, its orientation and the voxel
size to arbitrary values, or to select a different set of projections for a cardiac reconstruction. Thus the waiting
time for the volume reconstruction is removed. Therefore our method is esp. promising for applications such as
intra-operative CT and interventional CT.