Reduce Reconstruction Time Research Articles

3D deep encoder-decoder network for fluorescence molecular tomography.

Fluorescence molecular tomography (FMT) is a promising and noninvasive in vivo functional imaging modality. However, the quality of FMT reconstruction is limited by the simplified linear model of photon propagation. Here, an end-to-end three-dimensional deep encoder-decoder (3D-En-Decoder) network is proposed to improve the quality of FMT reconstruction. It directly establishes the nonlinear mapping relationship between the inside fluorescent source distribution and the boundary fluorescent signal distribution. Thus the reconstruction inaccuracy caused by the simplified linear model can be fundamentally avoided by the proposed network. Both numerical simulations and phantom experiments were carried out, and the results demonstrated that the 3D-En-Decoder network can greatly improve image quality and significantly reduce reconstruction time compared with conventional methods.

DRF-GRAPPA: A Parallel MRI Method with a Direct Reconstruction Filter

A novel multi-channel image reconstruction algorithm for parallel MRI is proposed to reconstruct a complex-valued image with both magnitude and phase information in a reduced computation time, so called the DRF-GRAPPA. The DRF-GRAPPA combines multichannel images by separately obtaining magnitude information and phase information. The method is applied to parallel MRI to reconstruct an image from under-sampled k-space data by using a direct reconstruction filter that can simultaneously estimate full k-space signals and combine multi-channel images. To obtain both magnitude and phase information, the DRF-GRAPPA provides two different direct reconstruction filters. The accuracy of the magnitude and phase information is validated through simulations. In addition, the experiment results of spherical water phantom imaging and in-vivo imaging demonstrate that the DRF-GRAPPA can successfully reconstruct images with both magnitude and phase information within a reduced reconstruction time.

Evaluation of respiratory-correlated 4D digital tomosynthesis imaging technique for image-guided radiation therapy

Recently, cone-beam computed tomography (CBCT) equipped with image-guided radiation therapy (IGRT) has been used to precisely identify the location of target lesion. However, the treatment accuracy for respiratory-sensitive regions is still low due to long scanning time and target motion, and the imaging dose is also relatively high compared to other imaging systems. These issues can be solved by using 4D IGRT with digital tomosynthesis (DT). Also, an application of graphics processing unit (GPU) computing can reduce reconstruction time. The purpose of this study was to develop a 4D DT imaging technique for IGRT and compare image quality between 3D DT and 4D DT. And, we also compared reconstruction time between CPU and GPU computing. Projections were separately obtained through 3 phases for simulating 4D DT imaging. We measured profile, normalized root-mean-square error (NRMSE), contrast-to-noise ratio (CNR), noise-power spectrum (NPS) and figure-of-merit (FOM). NRMSEs and CNRs of 4D DT images were averagely 0.56 and 0.74 times lower than those of 3D DT images, respectively. The NPSs of 4D DT images were higher than 3D DT images. The FOMs of 4D DT images were averagely 1.65 times higher than 3D DT images. The GPU computing reduced reconstruction time in 3D and 4D imaging by a factor of 0.11 and 0.56, respectively. As a results, the 4D DT imaging enables the high-precision IGRT for respiratory-sensitive regions, and the real-time 4D imaging would be possible by using the GPU computing.

Phase unwinding for dictionary compression with multiple channel transmission in magnetic resonance fingerprinting

PurposeMagnetic Resonance Fingerprinting reconstructions can become computationally intractable with multiple transmit channels, if the B1+ phases are included in the dictionary. We describe a general method that allows to omit the transmit phases. We show that this enables straightforward implementation of dictionary compression to further reduce the problem dimensionality. MethodsWe merged the raw data of each RF source into a single k-space dataset, extracted the transceiver phases from the corresponding reconstructed images and used them to unwind the phase in each time frame. All phase-unwound time frames were combined in a single set before performing SVD-based compression. We conducted synthetic, phantom and in-vivo experiments to demonstrate the feasibility of SVD-based compression in the case of two-channel transmission. ResultsUnwinding the phases before SVD-based compression yielded artifact-free parameter maps. For fully sampled acquisitions, parameters were accurate with as few as 6 compressed time frames. SVD-based compression performed well in-vivo with highly under-sampled acquisitions using 16 compressed time frames, which reduced reconstruction time from 750 to 25min. ConclusionOur method reduces the dimensions of the dictionary atoms and enables to implement any fingerprint compression strategy in the case of multiple transmit channels.

Low Overhead Architectures for OMP Compressive Sensing Reconstruction Algorithm

Orthogonal Matching Pursuit (OMP) is an important compressive sensing (CS) recovery and sparsity inducing algorithm, which has potential in various emerging applications ranging from wearable and mobile computing to real-time analytics processing on servers. Thus application aware OMP algorithm implementation is important. In this paper, we propose two different modifications to OMP algorithm named Thresholding technique for OMP (tOMP) and Gradient Descent OMP (GDOMP) to reduce hardware complexity of OMP algorithm. tOMP modifies identification stage of OMP algorithm to reduce reconstruction time and GDOMP modifies residual update phase to reduce chip area. To demonstrate reconstruction efficiency of proposed OMP modifications, we compare signal-to-reconstruction error rate (SRER), signal-to-noise ratio (PSNR), and Structural Similarity index (SSIM) of previously proposed matching pursuit algorithms such as Subspace Pursuit (SP), Look Ahead OMP (LAOMP), and OMP, with tOMP, and GDOMP. We implemented reconfigurable, parallel, and pipelined architectures for three algorithms including OMP, tOMP, and GDOMP which can reconstruct different data vector sizes ranging from 128 to 1024, on 65 nm CMOS technology operating at 1 V supply voltage. The post place and route analysis on area, power, and latency show that, tOMP requires 33% less reconstruction time, and GDOMP consumes 44% less chip area when compared to OMP ASIC implementation. Compared to previously published work, the proposed architectures achieve 2.1 times improvement in Area-Delay product (ADP) and consume 40% less energy.

Joint System and Algorithm Design for Computationally Efficient Fan Beam Coded Aperture X-Ray Coherent Scatter Imaging

In x-ray coherent scatter tomography, tomographic measurements of the forward scatter distribution are used to infer scatter densities within a volume. A radiopaque 2D pattern placed between the object and the detector array enables the disambiguation between different scatter events. The use of a fan beam source illumination to speed up data acquisition relative to a pencil beam presents computational challenges. To facilitate the use of iterative algorithms based on a penalized Poisson log-likelihood function, efficient computational implementation of the forward and backward models are needed. Our proposed implementation exploits physical symmetries and structural properties of the system and suggests a joint system-algorithm design, where the system design choices are influenced by computational considerations, and in turn lead to reduced reconstruction time. Computational-time speedups of approximately 146 and 32 are achieved in the computation of the forward and backward models, respectively. Results validating the forward model and reconstruction algorithm are presented on simulated analytic and Monte Carlo data.

Optimization of Distributed Multi-Node, Multi-GPU, Heterogeneous System for 3D Image Reconstruction in Electrical Capacitance Tomography

Abstract Electrical Capacitance Tomography is a non-invasive imaging technique, which allows visualization of the industrial processes interior and can be applied to many branches of the industry. Image reconstruction process, especially in case of 3D images, is a very time consuming task (when using classic processors and algorithms), which in turn leads to an unacceptable waiting time and currently limits the use of 3D Electrical Capacitance Tomography. Reconstruction using deterministic methods requires execution of many basic operations of linear algebra, such as matrix transposition, multiplication, addition and subtraction. In order to reach real-time reconstruction a 3D ECT computational subsystem must be able to transform capacitance data into images in a fraction of a second. By assuming, that many of the computations can be performed in parallel using modern, fast graphics processor and by altering the algorithms, time to achieve high quality image reconstruction will be shortened significantly. The research conducted while analysing ECT algorithms has also shown that, although dynamic development of GPU computational capabilities and its recent application for image reconstruction in ECT has significantly improved calculations time, in modern systems a single GPU is not enough to perform many tasks. Distributed Multi-GPU solutions can reduce reconstruction time to only a fraction of what was possible on pure CPU systems. Nevertheless performed tests clearly illustrate the need for further optimizations of previously developed algorithms.

Analysis of Application of Distributed Multi-Node, Multi-GPU Heterogeneous System for Acceleration of Image Reconstruction in Electrical Capacitance Tomography

Abstract 3D ECT provides a lot of challenging computational issues that have been reported in the past by many researchers. Image reconstruction using deterministic methods requires execution of many basic operations of linear algebra, such as matrix transposition, multiplication, addition and subtraction. In order to reach real-time reconstruction a 3D ECT computational subsystem has to be able to transform capacitance data into image in fractions of seconds. By assuming, that many of the computations can be performed in parallel using modern, fast graphics processor and by altering the algorithms time to achieve high quality image reconstruction will be shortened significantly. The research conducted while analysing ECT algorithms has also shown that, although dynamic development of GPU computational capabilities and its recent application for image reconstruction in ECT has significantly improved calculations time, in modern systems a single GPU is not enough to perform many tasks. Distributed Multi-GPU solutions can reduce reconstruction time to only a fraction of what was possible on pure CPU systems. Nevertheless performed tests clearly illustrate the need for developing a new distributed platform, which would be able to fully utilize the potential of the hardware. It has to take into account specific nature of computations in Multi-GPU systems.

Adaptive rendering with linear predictions

We propose a new adaptive rendering algorithm that enhances the performance of Monte Carlo ray tracing by reducing the noise, i.e., variance, while preserving a variety of high-frequency edges in rendered images through a novel prediction based reconstruction. To achieve our goal, we iteratively build multiple, but sparse linear models. Each linear model has its prediction window, where the linear model predicts the unknown ground truth image that can be generated with an infinite number of samples. Our method recursively estimates prediction errors introduced by linear predictions performed with different prediction windows, and selects an optimal prediction window minimizing the error for each linear model. Since each linear model predicts multiple pixels within its optimal prediction interval, we can construct our linear models only at a sparse set of pixels in the image screen. Predicting multiple pixels with a single linear model poses technical challenges, related to deriving error analysis for regions rather than pixels, and has not been addressed in the field. We address these technical challenges, and our method with robust error analysis leads to a drastically reduced reconstruction time even with higher rendering quality, compared to state-of-the-art adaptive methods. We have demonstrated that our method outperforms previous methods numerically and visually with high performance ray tracing kernels such as OptiX and Embree.

Analysis of application of distributed multi-node, multi-GPU heterogeneous system for acceleration of image reconstruction in Electrical Capacitance Tomography

Abstract 3D ECT provides a lot of challenging computational issues that have been reported in the past by many researchers. Image reconstruction using deterministic methods requires execution of many basic operations of linear algebra, such as matrix transposition, multiplication, addition and subtraction. In order to reach real-time reconstruction a 3D ECT computational subsystem has to be able to transform capacitance data into image in fractions of seconds. By assuming, that many of the computations can be performed in parallel using modern, fast graphics processor and by altering the algorithms time to achieve high quality image reconstruction will be shortened significantly. The research conducted while analysing ECT algorithms has also shown that, although dynamic development of GPU computational capabilities and its recent application for image reconstruction in ECT has significantly improved calculations time, in modern systems a single GPU is not enough to perform many tasks. Distributed Multi-GPU solutions can reduce reconstruction time to only a fraction of what was possible on pure CPU systems. Nevertheless performed tests clearly illustrate the need for developing a new distributed platform, which would be able to fully utilize the potential of the hardware. It has to take into account specific nature of computations in Multi-GPU systems.

FPGA-Based Embedded Signal Processing for 3-D Ultrasound Computer Tomography

A three-dimensional ultrasound computer tomography (3-D USCT) system for early breast cancer diagnosis is currently being developed at Institute for Data Processing and Electronics (IPE) at Karlsruhe Institute of Technology, Karlsruhe, Germany. A field-programmable gate array (FPGA)-based data acquisition (DAQ) system is used to acquire the measurement data. Over 3.5 million data sets are recorded for each breast, resulting in 20 GB of raw data. The data is subsequently transferred to a PC for time-consuming image reconstruction, which is in the order of days. A promising approach in reducing reconstruction time is mapping further computation to the FPGAs in the DAQ system. This paper covers the investigation of FPGA-based processing for a set of signal processing algorithms. In the best suited approach, processing time per data set is approximately 50 μs, occupying less than 30% of device resources.

Dual-energy Computed Tomography Imaging of the Aorta

There are 2 inseparable and complimentary technical advantages of dual-energy computed tomography (CT) imaging of the thoracic aorta. One advantage stems from the simultaneous availability of low and high peak kilovoltage (kVp) spectra data and, in particular, the benefits conferred by the improved conspicuity of iodinated contrast media at lower kVp CT imaging. This, in turn, permits improved aortic visualization or, alternatively, reduction in the volume or rate of contrast administration. Image noise at low kilovoltage does not appear to be a significant issue, with the backup availability of simultaneously acquired high kVp images a distinct advantage over single, low kVp imaging techniques. The second advantage of dual-energy CT imaging stems from the potential to calculate material-specific images derived mathematically from the simultaneous availability of attenuation measurements at 2 distinct energies. These material-specific data sets include virtual noncontrast images, virtual contrast, or "bone-subtracted" angiographic-like images. These techniques may confer significant advantages in the evaluation of patients with aortic disease, improving interpretation and reducing reconstruction time, while potentially reducing the need for, and associated radiation burden of, precontrast or postcontrast multiphasic imaging.

Nonlinear inverse reconstruction for real‐time MRI of the human heart using undersampled radial FLASH

A previously proposed nonlinear inverse reconstruction for autocalibrated parallel imaging simultaneously estimates coil sensitivities and image content. This work exploits this property for real-time MRI, where coil sensitivities need to be dynamically adapted to the conditions generated by moving objects. The development comprises (i) an extension of the nonlinear inverse algorithm to non-Cartesian k-space encodings, (ii) its implementation on a graphical processing unit to reduce reconstruction times, and (iii) the use of a convolution-based iteration, which considerably simplifies the graphical processing unit implementation compared to a gridding technique. The method is validated for real-time MRI of the human heart at 3 T using radio frequency-spoiled radial FLASH (pulse repetition time/echo time = 2.0/1.3 ms, flip angle 8 degrees). The results demonstrate artifact-free reconstructions from only 65-85 spokes, with 256 oversampled data points. Acquisition times of 130-170 ms resulted in 29-38 frames per second for sliding window reconstructions (factor 5). While offline reconstructions required 1-2 sec, real-time applications with modified parameters and slightly lower image quality were achieved within 90 ms per graphical processing unit.

Optimization of time-of-flight reconstruction on Philips GEMINI TF

The aim of this study is to optimize different parameters in the time-of-flight (TOF) reconstruction for the Philips GEMINI TF. The use of TOF in iterative reconstruction introduces additional variables to be optimized compared to conventional PET reconstruction. The different parameters studied are the TOF kernel width, the kernel truncation (used to reduce reconstruction time) and the scatter correction method. These parameters are optimized using measured phantom studies. All phantom studies were acquired with a very high number of counts to limit the effects of noise. A high number of iterations (33 subsets and 3 iterations) was used to reach convergence. The figures of merit are the uniformity in the background, the cold spot recovery and the hot spot contrast. As reference results we used the non-TOF reconstruction of the same data sets. It is shown that contrast recovery loss can only be avoided if the kernel is extended to more than 3 standard deviations. To obtain uniform reconstructions the recommended scatter correction is TOF single scatter simulation (SSS). This also leads to improved cold spot recovery and hot spot contrast. While the daily measurements of the system show a timing resolution in the range of 590–600 ps, the optimal reconstructions are obtained with a TOF kernel full-width at half-maximum (FWHM) of 650–700 ps. The optimal kernel width seems to be less critical for the recovered contrast but has an important effect on the background uniformity. Using smaller or wider kernels results in a less uniform background and reduced hot and cold contrast recovery. The different parameters studied have a large effect on the quantitative accuracy of the reconstructed images. The optimal settings from this study can be used as a guideline to make an objective comparison of the gains obtained with TOF PET versus PET reconstruction.

3D reconstruction of the spine from biplanar X-rays using parametric models based on transversal and longitudinal inferences

Reconstruction methods from biplanar X-rays provide 3D analysis of spinal deformities for patients in standing position with a low radiation dose. However, such methods require an important reconstruction time and there is a clinical need for fast and accurate techniques. This study proposes and evaluates a novel reconstruction method of the spine from biplanar X-rays. The approach uses parametric models based on longitudinal and transversal inferences. A first reconstruction level, dedicated to routine clinical use, allows to get a fast estimate (reconstruction time: 2 min 30 s) of the 3D reconstruction and accurate clinical measurements. The clinical measurements precision (evaluated on asymptomatic subjects, moderate and severe scolioses) was between 1.2° and 5.6°. For a more accurate 3D reconstruction (complex pathologies or research purposes), a second reconstruction level can be obtained within a reduced reconstruction time (10 min) with a fine adjustment of the 3D models. The mean shape accuracy in comparison with CT-scan was 1.0 mm. The 3D reconstruction method precision was 1.8 mm for the vertebrae position and between 2.3° and 3.9° for the orientation. With a reduced reconstruction time, an improved accuracy and precision and a method proposing two reconstruction levels, this approach is efficient for both clinical routine uses and research purposes.

Vascular tree reconstruction with discrete tomography: intensity based camera correction for 3D reconstruction

This paper is concerned with the reconstruction of vascular trees from few projections using discrete tomography. However, its computational cost is high and it lacks robustness when the data are inconsistent. We improve robustness by incorporating an intensity-based camera-correction method. The proposed approach is also capable of handling small motion artifacts by modeling them as repositionings of a virtual X-ray camera. We also present a parallel implementation which substantially reduces reconstruction time. We propose a data-driven reduction of positional inconsistencies by minimizing the reconstruction residual to increase the robustness. Inspired by motion compen-sation algorithms in SPECT imaging, we combine an intensity-based 2D/3D-registration method with itera-tive reconstruction methods. Our objective is the robust vascular-tree reconstruction from positionally inconsistent data. The speed of the reconstruction is substantially increased by a volume-splitting scheme that allows parallel processing. Vascular trees in the liver can be accurately reconstructed from few positionally inconsistent projections using digitally reconstructed radiographs. We have tested the proposed method on synthetic projection data and on objects imaged with a new robotized C-arm. We measured a decrease in the average reconstruction residual of about 13% for real data compared to projection data without preprocessing. Over 4,600 reconstruction experiments were conducted to evaluate the speed-up obtained when employing the volume-splitting scheme. Reconstruction time decreased linearly with increased number of processor-cores, both for real and synthetic data. The proposed method reduces inconsistencies caused by positioning errors and small motion artifacts. No prior segmentation or detection of correspondences between projections is necessary, because all algorithms are intensity-based. As a result, the proposed method allows for robust, high-quality reconstructions, while reducing radiation dose substantially.

Comparison of multi-ray and point-spread function based resolution recovery methods in pinhole SPECT reconstruction

Statistical reconstruction methods allow resolution recovery in tomographic reconstruction. Even though resolution recovery has the potential to improve overall image quality, pinhole SPECT images are still often reconstructed using simplified models of the acquisition geometry in order to reduce reconstruction time. This paper investigates the benefits of two resolution recovery methods, multi-ray and point-spread function based, in pinhole SPECT by comparing them to uncorrected reconstruction. Resolution recovery was incorporated into ordered subsets expectation maximization reconstruction algorithm. The first of the correction methods used a simple but very fast multiple projection ray approach, whereas the second, much slower, method modelled the acquisition geometry more accurately using the analytical point-spread function of the pinhole collimator. Line source, Jaszczak and contrast phantom studies were performed and used for comparison. Resolution recovery improved resolution, contrast and visual quality of the images when compared to reconstructions without it. The method based on the point-spread function performed slightly better, but was almost 50 times slower than the much simpler multi-ray approach. The multiple projection ray approach is a promising method for very fast and easy resolution recovery in pinhole SPECT. It has a profound effect on image quality and can markedly improve the resolution-sensitivity trade-off.

Evaluation of accelerated iterative x-ray CT image reconstruction using floating point graphics hardware

Statistical reconstruction methods offer possibilities to improve image quality as compared with analytical methods, but current reconstruction times prohibit routine application in clinical and micro-CT. In particular, for cone-beam x-ray CT, the use of graphics hardware has been proposed to accelerate the forward and back-projection operations, in order to reduce reconstruction times. In the past, wide application of this texture hardware mapping approach was hampered owing to limited intrinsic accuracy. Recently, however, floating point precision has become available in the latest generation commodity graphics cards. In this paper, we utilize this feature to construct a graphics hardware accelerated version of the ordered subset convex reconstruction algorithm. The aims of this paper are (i) to study the impact of using graphics hardware acceleration for statistical reconstruction on the reconstructed image accuracy and (ii) to measure the speed increase one can obtain by using graphics hardware acceleration. We compare the unaccelerated algorithm with the graphics hardware accelerated version, and for the latter we consider two different interpolation techniques. A simulation study of a micro-CT scanner with a mathematical phantom shows that at almost preserved reconstructed image accuracy, speed-ups of a factor 40 to 222 can be achieved, compared with the unaccelerated algorithm, and depending on the phantom and detector sizes. Reconstruction from physical phantom data reconfirms the usability of the accelerated algorithm for practical cases.

A Monte Carlo and physical phantom evaluation of quantitative In-111 SPECT

Accurate estimation of the 3D in vivo activity distribution is important for dose estimation in targeted radionuclide therapy (TRT). Although SPECT can potentially provide such estimates, SPECT without compensation for image degrading factors is not quantitatively accurate. In this work, we evaluated quantitative SPECT (QSPECT) reconstruction methods that include compensation for various physical effects. Experimental projection data were obtained using a GE VH/Hawkeye system and an RSD torso phantom. Known activities of In-111 chloride were placed in the lungs, liver, heart, background and two spherical compartments with inner diameters of 22 mm and 34 mm. The 3D NCAT phantom with organ activities based on clinically derived In-111 ibritumomab tiuxetan data was used for the Monte Carlo (MC) simulation studies. Low-noise projection data were simulated using previously validated MC simulation methods. Fifty sets of noisy projections with realistic count levels were generated. Reconstructions were performed using the OS-EM algorithm with various combinations of attenuation (A), scatter (S), geometric response (G), collimator–detector response (D) and partial volume compensation (PVC). The QSPECT images from the various combinations of compensations were evaluated in terms of the accuracy and precision of the estimates of the total activity in each organ. For experimental data, the errors in organ activities for ADS and PVC compensation were less than 6.5% except the smaller sphere (−11.9%). For the noisy simulated data, the errors in organ activity for ADS compensation were less than 5.5% except the lungs (20.9%) and blood vessels (15.2%). Errors for other combinations of compensations were significantly (A, AS) or somewhat (AGS) larger. With added PVC, the error in the organ activities improved slightly except for the lungs (11.5%) and blood vessels (3.6%) where the improvement was more substantial. The standard deviation/mean ratios were all less than 1.5%. We conclude that QSPECT methods with appropriate compensations provided accurate In-111 organ activity estimates. For the collimator used, AGS was almost as good as ADS and may be preferable due to the reduced reconstruction time. PVC was important for small structures such as tumours or for organs in close proximity to regions with high activity. The improved quantitative accuracy from QSPECT methods has the potential for improving organ dose estimations in TRT.

Statistical image reconstruction for transmission tomography using relaxed ordered subset algorithms

Statistical reconstruction methods offer possibilities for improving image quality as compared to analytical methods, but current reconstruction times prohibit routine clinical applications in x-ray computed tomography (CT). To reduce reconstruction times, we have applied (under) relaxation to ordered subset algorithms. This enables us to use subsets consisting of only single projection angle, effectively increasing the number of image updates within an entire iteration. A second advantage of applying relaxation is that it can help improve convergence by removing the limit cycle behaviour of ordered subset algorithms, which normally do not converge to an optimal solution but rather a suboptimal limit cycle consisting of as many points as there are subsets. Relaxation suppresses the limit cycle behaviour by decreasing the stepsize for approaching the solution. A simulation study for a 2D mathematical phantom and three different ordered subset algorithms shows that all three algorithms benefit from relaxation: equal noise-to-resolution trade-off can be achieved using fewer iterations than the conventional algorithms, while a lower minimal normalized mean square error (NMSE) clearly indicates a better convergence. Two different schemes for setting the relaxation parameter are studied, and both schemes yield approximately the same minimal NMSE.