Tomographic microscopy using synchrotron radiation provides high-resolution structure details on the scale of
microns. The field of view (FOV) of the microscopy system, however, is usually limited by the detector size.
For example, a typical CCD camera used for data acquisition is of size 2048 by 2048. In many cases this CCD
camera is not large enough to provide complete information required for accurate reconstruction, and the local
tomography problem hereby arises. On the other hand, the huge dataset generated by tomographic microscopy
asks for a highly efficient solution with no a priori information necessary. A new padding scheme is therefore
proposed for the local tomography issue. It first pads the projection data using the boundary value inside the
FOV, which is specified by the detector size, followed by a zero-value padding to 1.5 times the FOV length.
The boundary-value padding removes the energy deposition and cupping artifact in reconstruction results from
local tomography, while the zero-value padding reduces the drift of the intensity values caused by fully boundary
padding. The combination of two padding schemes keeps advantages of fully zero-value padding and fully
boundary-value padding, while avoiding their disadvantages. Quantitative analysis using synthetic data shows
that the proposed method outperforms fully zero-value padding and fully boundary-value padding in terms of
accuracy and ease for post processing. Experimental results for real data are also provided to demonstrate the
effectiveness of the proposed method.
A typical iterative CT reconstruction using SART involves ray-driven forward projection and voxel-driven back-
ward projection. Bilinear interpolation is usually applied on image data for forward projection, and linear
interpolation is usually applied on projection data for backward projection, when both data are represented
using discrete samples in 2D fan-beam geometry. The applied interpolations, however, may affect the spatial
resolution, bias and noise properties of the reconstruction. A basis function (such as blob and spline) is therefore
applied to formulate a continuous model for the image data to reduce bias. In this paper we propose to apply
the blob representation on the projection data and explore its effectiveness. In this way we use continuous model
for the data of projection difference during backward projection, and we avoid the linear interpolation in this
process. Experimental results show that the proposed scheme is able to provide higher spatial resolution than
linear interpolation, while introducing more local variations in the reconstruction. However, the introduced local
variations may be reduced with the combination of total variation (TV) minimization. The proposed scheme is
therefore able to provide improved spatial resolution while keeping low local variations in reconstructions.
Ring artifacts are common in tomography reconstructions. These artifacts may arise, in microtomography
using synchrotron radiation, from dead pixels in CCD detectors, damaged scintillator
screens and instabilities of the synchrotron beam. Ring artifact removal is therefore an important
task. We apply an image inpainting algorithm on the sinogram data to reduce ring artifacts in
reconstructions. The applied image inpainting algorithm is advantageous in two aspects. First,
the proposed method is local, and it affects only the user-specified corrupted region. Second,
it utilizes the directional information of level lines, as well as the intensity information, in the
sinogram data. Experimental results show that the proposed method is able to compensate both
the missing directional information and the distorted intensity information, which reduces ring
artifacts in reconstructions. An improvement in reconstruction image qualities is achieved for
micro-tomography using the proposed method. A comparison shows that the proposed method
may outperform existing methods in terms of reconstruction accuracy.
Determination of the rotation axis position for tomographic projection images is critical to perform
an accurate reconstruction. Rotational centers in micro-tomography may shift by several
microns after the initial calibration due to various factors such as temperature variation, sample
system stability and sample loading procedures. Automatic detection of rotational centers after
data acquisition is therefore crucial for accurate and efficient reconstructions, and it is commonly
implemented at various synchrotron facilities. We propose to implement a reliable cross correlation
method on the projections of 0 and 180 degree to automatically re-align the rotation axis at data
collection time. For this purpose, several issues, such as the flat-field correction for the imaging
system and the irregular data near projection boundaries, are handled to increase the stability
achieving subpixel alignments. The method is shown from experimental results to be accurate,
efficient and stable. The results from automatic detections are mostly within one pixel difference
from manual/operator detection results. Following the data collection we developed an automatic
sub-pixel rotational centering method. Intermediate results from this final process are generated for
user inspection. The proposed method is able to detect rotational center shifts within 7 seconds for
high-resolution projections of size 2048×2048. It is shown to be stable for static samples in complicated
cases. GPU is utilized to fasten the cross correlation computation in the space domain, which
achieves about 10 times speedup. The proposed method fits seamlessly into the current framework
of beamline 2-BM at the Advanced Photon Source, Argonne National Laboratory. It may save
5 minutes for partial reconstructions and 5-10 minutes for manual detections without sacrificing
Iterative CT reconstruction methods have advantages over analytical reconstruction methods because
of their robustness to both noise and incomplete projection data, which have great potential
for dose reduction in real applications. The SART algorithm, which is one of the well-established
iterative reconstruction methods, has been examined extensively, and GPU has been applied to
improve their efficiency. Although it has been proved that SART may globally converge, its convergence
is very slow, especially after the first several iterations. Hundreds of iterations may be
needed for accurate reconstruction. This slow convergence requires heavy data transfer between
global memory and texture memory inside GPU. Therefore, preconditioned conjugate gradient (CG)
method, which converges much faster than SART, may be combined with SART for better performance.
Since CG is sensitive to initialization, the reconstruction results from SART after a few
iterations may be used as the initialization for CG. Preliminary experimental results on CPU show
that this framework converges much faster than SART and CG, which demonstrates its potential
in real applications.
A precise knowledge of the geometry information of a cone-beam CT is required for high quality
image reconstruction. In some applications, however, the acquisition geometry is either not well
characterized or not repeatable, for example, in the case of gantry vibration and patient motion.
An offline correction using a calibration pattern and an online calibration using an external tracking
system may be used to measure and correct CT geometry parameters during reconstruction.
Both approaches have limitations though. A new method is proposed in this paper to estimate
pose parameters using the acquired cone-beam projection data only. During the pose estimation
process each 2D projection data is registered to the 3D volume reconstructed from the current,
inaccurate pose estimate. Pose parameters are then corrected incrementally using the registration
results. Applying this 2D-3D registration method to the FDK reconstruction method, we are able
to estimate rotational parameters to within an average total angular deviation of 0.5 degrees, and
center-of-rotation to an average of 0.02% of the source-to-detector distance (SID) in the detector
plane. The image quality of CT reconstructions is comparable to those using exact geometry.
This paper presents a simulation study on iterative reconstruction methods for helical cone-beam
CT using graphics hardware. While analytic methods have been proposed to achieve an exact
reconstruction for helical cone-beam CT, these methods may not perform well for projection noise
and imaging geometry variations, especially for spiral variations. Iterative methods such as SART
are advantageous in this aspect. Since SART is computationally intense, graphics hardware (GPU)
may be utilized to handle the computations and increase the efficiency of SART. GPU SART
reconstruction results are presented for spiral cone-beam CT, and these results are compared with
the Katsevich reconstructions in presence of projection noise and spiral variations. The comparison
shows that SART is accurate, efficient, and more robust to projection noise and spiral variations
than the Katsevich method.
3D computed tomography has been extensively studied and widely used in modern society.
Although most manufacturers choose the filtered backprojection algorithm (FBP) for its accuracy
and efficiency, iterative reconstruction methods have a significant potential to provide superior
performance for incomplete, noisy projection data. However, iterative methods have a high
computational cost, which hinders their practical use. Furthermore, regularization is usually
required to reduce the effects of noise. In this paper, we analyze the use of the Simultaneous
Algebraic Reconstruction Technique (SART) with total variation (TV) regularization.
Additionally, graphics hardware is utilized to increase the speed of SART. NVIDIA's GPU and
Compute Unified Device Architecture (CUDA) comprise the core of our computational platform.
GPU implementation details, including ray-based forward projection and voxel-based
backprojection are illustrated. Experimental results for high-resolution synthetic and real data are
provided to demonstrate the accuracy and efficiency of the proposed framework.
Computed tomography (CT) has been extensively studied and widely used for a variety of medical
applications. The reconstruction of 3D images from a projection series is an important aspect
of the modality. Reconstruction by filtered backprojection (FBP) is used by most manufacturers
because of speed, ease of implementation, and relatively few parameters. Iterative reconstruction
methods have a significant potential to provide superior performance with incomplete or
noisy data, or with less than ideal geometries, such as cone-beam systems. However, iterative
methods have a high computational cost, and regularization is usually required to reduce the
effects of noise. The simultaneous algebraic reconstruction technique (SART) is studied in this
paper, where the Feldkamp method (FDK) for filtered back projection is used as an initialization
for iterative SART. Additionally, graphics hardware is utilized to increase the speed of SART
implementation. Nvidia processors and compute unified device architecture (CUDA) form the
platform for GPU computation. Total variation (TV) minimization is applied for the regularization
of SART results. Preliminary results of SART on 3-D Shepp-Logan phantom using using
TV regularization and GPU computation are presented in this paper. Potential improvements
of the proposed framework are also discussed.