Iterative Reconstruction Methods Research Articles

Development of a Multiuse Human-Scale Single-Ring OpenPET System

We developed a human-scale single-ring OpenPET (SROP) system, which had an open space allowing us access to the subject during measurement. The SROP system consisted of 160 4-layer depth-of-interaction detectors. The open space with the axial width of 430 mm was achieved with the ring axial width of 214 mm and the ring inner diameter of 660 mm. The detectors were axially shifted to each other so that the detector ring was aligned along a plane horizontally tilted by 45° against the axial direction. The system was developed as a mobile scanner to be used not only in clinical positron emission tomography (PET) rooms but also in charged-particle therapy treatment rooms as well as animal experiment rooms. Almost uniform spatial resolution better than 3 mm throughout the entire field of view (FOV) was realized with an iterative image reconstruction method. Peak absolute sensitivity was 3.1%, and there was a region with sensitivity better than 0.8% for a length of more than 700 mm. An in-beam imaging experiment conducted at the heavy ion medical accelerator in Chiba showed that the system was operable even at the highest beam intensity available for heavy-ion therapy. In addition, we conducted entire-body monkey dynamic imaging utilizing the long region inside the gantry by positioning a monkey along the direction having the longest FOV tilted by 45° against the axial direction. We concluded the developed system has a capability to realize versatile PET applications by utilizing its wide-open space and mobility in addition to high spatial resolution with sufficiently good sensitivity. -9mm]Please consider rephrasing the sentence “We concluded the developed system” for clarity.

Artificial intelligence in image reconstruction: The change is here.

Innovations in CT have been impressive among imaging and medical technologies in both the hardware and software domain. The range and speed of CT scanning improved from the introduction of multidetector-row CT scanners with wide-array detectors and faster gantry rotation speeds. To tackle concerns over rising radiation doses from its increasing use and to improve image quality, CT reconstruction techniques evolved from filtered back projection to commercial release of iterative reconstruction techniques, and recently, of deep learning (DL)-based image reconstruction. These newer reconstruction techniques enable improved or retained image quality versus filtered back projection at lower radiation doses. DL can aid in image reconstruction with training data without total reliance on the physical model of the imaging process, unique artifacts of PCD-CT due to charge sharing, K-escape, fluorescence x-ray emission, and pulse pileups can be handled in the data-driven fashion. With sufficiently reconstructed images, a well-designed network can be trained to upgrade image quality over a practical/clinical threshold or define new/killer applications. Besides, the much smaller detector pixel for PCD-CT can lead to huge computational costs with traditional model-based iterative reconstruction methods whereas deep networks can be much faster with training and validation. In this review, we present techniques, applications, uses, and limitations of deep learning-based image reconstruction methods in CT.

Improving image quality in fast, time-resolved micro-CT by weighted back projection

Time-resolved micro-CT is an increasingly powerful technique for studying dynamic processes in materials and structures. However, it is still difficult to study very fast processes with this technique, since fast scanning is typically associated with high image noise levels. We present weighted back projection, a technique applicable in iterative reconstruction methods using two types of prior knowledge: (1) a virtual starting volume resembling the sample, for example obtained from a scan before the dynamic process was initiated, and (2) knowledge on which regions in the sample are more likely to undergo the dynamic process. Therefore, processes on which this technique is applicable are preferably occurring within a static grid. Weighted back projection has the ability to handle small errors in the prior knowledge, while similar 4D micro-CT techniques require the prior knowledge to be exactly correct. It incorporates the prior knowledge within the reconstruction by using a weight volume, representing for each voxel its probability of undergoing the dynamic process. Qualitative analysis on a sparse subset of projection data from a real micro-CT experiment indicates that this method requires significantly fewer projection angles to converge to a correct volume. This can lead to an improved temporal resolution.

3-D photoacoustic tomography reconstruction with deep learning for vasculature imaging

Photoacoustic tomography (PAT) is a hybrid imaging modality capable of acquiring high contrast and resolution images of optical absorption at depths greater than traditional optical imaging techniques. Practical considerations with instrumentation limit the number of acoustic sensors and their “view” of the imaging target, which result in image reconstruction artifacts degrading image quality. Iterative reconstruction methods can be used to improve image quality but are computationally expensive, especially for 3-D PAT over large imaging volumes. Deep learning has emerged as an efficient alternative and is capable of achieving state-of-the-art performance. In this work, we compare the 3-D fully dense UNet convolutional neural network with the widely used time reversal reconstruction method. Simulated acoustic data was generated using the K-wave MATLAB toolbox (225 detectors, planar geometry). A public database containing 50 patient's whole-lung CT scans was used to create training (n = 40) and testing (n = 10) data. The training and testing datasets are comprised of 500 and 50 lung vasculature volumes sampled from their respective whole-lung CT scans. Using the structural similarity index, the proposed deep learning method (0.87 ± 0.04) is shown to be superior to time reversal (0.38 ± 0.11).

Increasing the Performance of the Iterative Computed Tomography Image Reconstruction Algorithms

Computed tomography (CT) imaging is an important diagnostic tool. CT imaging facilitates the internal rendering of a scanned object by measuring the attenuation of beams of X-ray radiation. CT employs a mathematical technique of image reconstruction; those techniques are classified as; analytical and iterative. The iterative reconstruction (IR) methods have been proven to be superior over the analytical methods, but due to their prolonged reconstruction time, those methods are excluded from routine use in clinical applications. In this paper the reconstruction time of an IR algorithm is minimized through the employment of an adaptive region growing segmentation method that focuses the image reconstruction process on a specified region, thus ignoring unwanted pixels that increase the computation time. This method is tested on the iterative algebraic reconstruction technique (ART) algorithm. Some phantom images are used in this paper to demonstrate the effects of the segmentation process. The simulation results are executed using MATLAB (version R2018b) programming language, and a computer system with the following specifications: CPU core i7 (2.40 GHz) for processing. Simulation results indicate that this method will reduce the reconstruction time of the iterative algorithms, and will enhance the quality of the reconstructed image.

Development of ZJU High-Spectral-Resolution Lidar for Aerosol and Cloud: Extinction Retrieval

The retrieval of the extinction coefficients of aerosols and clouds without assumptions is the most important advantage of the high-spectral-resolution lidar (HSRL). The standard method to retrieve the extinction coefficient from HSRL signals depends heavily on the signal-to-noise ratio (SNR). In this work, an iterative image reconstruction (IIR) method is proposed for the retrieval of the aerosol extinction coefficient based on HSRL data, this proposed method manages to minimize the difference between the reconstructed and raw signals based on reasonable estimates of the lidar ratio. To avoid the ill-posed solution, a regularization method is adopted to reconstruct the lidar signals in the IIR method. The results from Monte-Carlo (MC) simulations applying both standard and IIR methods are compared and these comparisons demonstrate that the extinction coefficient and the lidar ratio retrieved by the IIR method have smaller root mean square error (RMSE) and relative bias values than the standard method. A case study of measurements made by Zhejiang University (ZJU) HSRL is presented, and their results show that the IIR method not only obtains a finer structure of the aerosol layer under the condition of low SNR, but it is also able to retrieve more reasonable values of the lidar ratio.

A deep unrolling network inspired by total variation for compressed sensing MRI

Compressed Sensing theory breaks through the limitation of the Nyquist sampling law and provides theoretical support for accelerating the imaging process of MRI while reconstructing high-quality images. It can use less sampling data through the reconstructed algorithm to restore the original signal. Reconstruction time and reconstruction algorithms play important roles in compressed sensing. However, it is difficult to apply the iterative reconstruction method in clinical applications because of the long reconstruction time. In addition, hand-crafted image prior which is widely used in traditional iterative method lacks of adaptivity for MRI reconstruction. In this paper, inspired by the total variation model, we propose a novel deep network for CSMRI, dubbed as TV-Inspired Network (TVINet), which incorporates the deep priors into the traditional iterative algorithm. We apply a general compressed sensing reconstruction framework inspired by TV regularization and solve it by combines the iterative method with deep learning. The design of TVINet comes from the inference process on the basis of Primal Dual Hybrid Gradient algorithm, which makes the network has preferable interpretability. The linear operator and regularization term are learned from the training dataset by using a convolution network. The proposed approach trains the network end-to-end and reconstructs the desired image directly from the under-sampling data. Experimental results show that the proposed method has better PSNR or SSIM than state-of-the-art methods, and maintains more complex and fine details in the reconstructed image.

Passive gamma emission tomography with ordered subset expectation maximization method

Gamma-ray emission tomography (GET) was developed as a potential verification tool to visualize the passive gamma-ray emitter sources of fuel rods. In GET, maximum likelihood–expectation maximization (MLEM) was employed as an iterative reconstruction method. However, as convergence iteration in the algorithm is proportional to the pixel size, convergence is slow and the calculation cost for practical application is high. Therefore, an ordered subset expectation maximization method (OSEM) was used, and the rod-wise relative gamma-ray emitter distribution of a BWR 10 × 10 mock-up fuel assembly was reconstructed to evaluate the performance of the OSEM. The OSEM enabled reconstruction comparable to that of MLEM with an effective decrease in the number of iterations.

Quantitative T2 mapping using accelerated 3D stack-of-spiral gradient echo readout

PurposeTo develop a rapid T2 mapping protocol using optimized spiral acquisition, accelerated reconstruction, and model fitting. Materials and methodsA T2-prepared stack-of-spiral gradient echo (GRE) pulse sequence was applied. A model-based approach joined with compressed sensing was compared with the two methods applied separately for accelerated reconstruction and T2 mapping. A 2-parameter-weighted fitting method was compared with 2- or 3-parameter models for accurate T2 estimation under the influences of noise and B1 inhomogeneity. The performance was evaluated using both digital phantoms and healthy volunteers. Mitigating partial voluming with cerebrospinal fluid (CSF) was also tested. ResultsSimulations demonstrates that the 2-parameter-weighted fitting approach was robust to a large range of B1 scales and SNR levels. With an in-plane acceleration factor of 5, the model-based compressed sensing-incorporated method yielded around 8% normalized errors compared to references. The T2 estimation with and without CSF nulling was consistent with literature values. ConclusionThis work demonstrated the feasibility of a T2 quantification technique with 3D high-resolution and whole-brain coverage in 2–3 min. The proposed iterative reconstruction method, which utilized the model consistency, data consistency and spatial sparsity jointly, provided reasonable T2 estimation. The technique also allowed mitigation of CSF partial volume effect.

Iterative versus non-iterative image reconstruction methods for sparse magnetic resonance imaging.

Magnetic resonance imaging (MRI) using under-sampled k-space data is a common method to shorten the imaging time. Iterative Bayesian algorithms are usually used for its image reconstruction. This paper compares an iterative Bayesian image reconstruction method that uses both spatial and temporal constraints and a non-iterative reconstruction algorithm that does not use temporal constraints. Three patient studies are performed. It is interesting to notice that the images reconstructed by the iterative Bayesian algorithm may introduce more bias than the non-iterative algorithm, even though the images provided by the iterative Bayesian algorithm look less noisy. The bias can be reduced by decreasing the influence of the temporal constraints.

Application of sigmoidal optimization to reconstruct nuclear medicine image: Comparison with filtered back projection and iterative reconstruction method

High levels for noise and a loss of true signal make the quantitative interpretation of nuclear medicine (NM) images difficult. An application of profile optimization using a sigmoidal function in this study was used to acquire the NM images with high quality. And the images were acquired by using three kinds of reconstruction method using each same sinogram: a standard filtered back-projection (FBP), an iterative reconstruction (IR) technique, and the sigmoidal function profile optimization (SFPO). Comparison of image according to reconstruction method was performed to show a superiority of the SFPO for imaging. The images reconstructed by using the SFPO showed an average of 1.49 times and of 1.17 times better in contrast than the results obtained using the standard FBP and the IR technique, respectively. Higher signal to noise ratios were obtained as an average of 12.30 times and of 3.77 times than results obtained using the standard FBP and the IR technique, respectively. This study confirms that reconstruction with SFPO (vs FBP and vs IR) can lead to better lesion detectability and characterization with noise reduction. It can be developed for future reconstruction technique for the NM imaging.

Iterative Model-Based Beamforming for High Dynamic Range Applications.

Clutter produced using bright acoustic sources can obscure weaker acoustic targets, degrading the quality of the image in scenarios with high dynamic ranges. Many adaptive beamformers seek to improve image quality by reducing these sidelobe artifacts, generating a boost in contrast ratio or contrast-to-noise ratio. However, some of these beamformers inadvertently introduce a dark region artifact in place of the strong clutter, a situation that occurs when both clutter and the underlying signal of interest are removed. We introduce the iterative aperture domain model image reconstruction (iADMIRE) method that is designed to reduce clutter while preserving the underlying signal. We compare the contrast ratio dynamic range (CRDR) of iADMIRE to several other adaptive beamformers plus delay-and-sum (DAS) to quantify the accuracy and reliability of the reported measured contrast for each beamformer over a wide range of contrast levels. We also compare all beamformers in the presence of bright targets ranging from 40 to 120 dB to observe the presence of sidelobes. In cases with no added reverberation clutter, iADMIRE had a CRDR of 75.6 dB when compared with the next best method DAS with 60.8 dB. iADMIRE also demonstrated the best performance for levels of reverberation clutter up to 0-dB signal-to-clutter ratio. Finally, iADMIRE restored underlying speckle signal in dark artifact regions while suppressing sidelobes in bright target cases up to 100 dB.

Effect of X-ray beam energy and image reconstruction technique on computed tomography numbers of various tissue equivalent materials

IntroductionComputed tomography (CT) numbers are used in radiological diagnosis, attenuation correction and radiotherapy treatment planning. Modern CT scanners use iterative reconstruction methods instead of the traditional filtered back projection (FBP). Hence, the investigation of CT number accuracy with image reconstruction techniques and X-ray tube potential (kVp) used in CT is warranted. The aim of this study is to evaluate the effect of Sinogram Affirmed Iterative Reconstruction (SAFIRE) Technique and image acquisition at different tube potentials on CT numbers of different tissue equivalent materials. MethodsImages of the Computerised Imaging Reference System Model 062M Electron Density Phantom were acquired at different tube potentials and reconstructed using FBP and different strengths of SAFIRE. Average CT numbers, in circular regions of interest, and their standard deviations were used to investigate any dependence of CT numbers on tube potentials and/or image reconstruction technique using non-parametric statistical tests with p-values set at 0.05. ResultsStatistically significant differences in CT numbers were not observed (p > 0.091) between the different image reconstruction techniques. CT number of bone equivalent materials increased significantly (p < 0.015), by up to 400 Hounsfield Units, when tube potential was decreased. Such extent of CT number change over the tube potentials range used in this study may influence diagnostic outcomes in lung nodule, contrast enhanced and calcium score studies. For all other tissue equivalent materials, the CT number did not change significantly for different tube potentials. Linear relationship was observed between CT numbers and electron densities. ConclusionThe study concludes that the CT numbers of all tissues did not change significantly with image reconstruction methods. However, the CT numbers of bone equivalent materials increased with decreasing tube potentials, which may result in misrepresentation of clinical information obtained. Implications for practiceWhen CT images are used to extract quantitative parameters such as calcium score, to characterise lung nodules and contrast enhanced structures, the kVp used for image acquisition should be carefully selected to avoid any misrepresentation of clinical information.

FPGA Acceleration for 3-D Low-Dose Tomographic Reconstruction

X-ray computed tomography (CT) is commonly used to obtain vivo images to characterize diseases but results in radiation exposure to patients. Low-dose CT (LDCT) provides CT images of clinical quality with reduced cumulative radiation dose. Iterative image reconstruction methods with effective regularization are used for LDCT but generally require more computing resources and induce higher computational load than the conventional filtered backprojection (FBP) methods. The high computational demand of the iterative reconstruction (IR) with notably increased reconstruction time precludes its routine clinical application. In this work, we focus on the FPGA acceleration of a compute-intensive full IR (full-IR) algorithm based on the Mumford–Shah regularization. At the algorithmic level, we propose a beam-based asynchronous update algorithm to reduce the computational cost and alleviate the conflicts. At the hardware-level, we first present pipeline-friendly optimization for the original algorithm to increase the computation throughput. We then apply the LDCT-specific tiling strategy to improve the data reuse rate. The experimental results show that our implementation takes 8.5 min to reconstruct a typical physical phantom with the image quality comparable with the vendor’s result. The FPGA implementation achieves $11.6\times $ throughput against the state-of-the-art GPU version.

Reconstruction of Compressed-sensing MR Imaging Using Deep Residual Learning in the Image Domain.

Purpose:A deep residual learning convolutional neural network (DRL-CNN) was applied to improve image quality and speed up the reconstruction of compressed sensing magnetic resonance imaging. The reconstruction performances of the proposed method was compared with iterative reconstruction methods.Methods:The proposed method adopted a DRL-CNN to learn the residual component between the input and output images (i.e., aliasing artifacts) for image reconstruction. The CNN-based reconstruction was compared with iterative reconstruction methods. To clarify the reconstruction performance of the proposed method, reconstruction experiments using 1D-, 2D-random under-sampling and sampling patterns that mix random and non-random under-sampling were executed. The peak-signal-to-noise ratio (PSNR) and the structural similarity index (SSIM) were examined for various numbers of training images, sampling rates, and numbers of training epochs.Results:The experimental results demonstrated that reconstruction time is drastically reduced to 0.022 s per image compared with that for conventional iterative reconstruction. The PSNR and SSIM were improved as the coherence of the sampling pattern increases. These results indicate that a deep CNN can learn coherent artifacts and is effective especially for cases where the randomness of k-space sampling is rather low. Simulation studies showed that variable density non-random under-sampling was a promising sampling pattern in 1D-random under-sampling of 2D image acquisition.Conclusion:A DRL-CNN can recognize and predict aliasing artifacts with low incoherence. It was demonstrated that reconstruction time is significantly reduced and the improvement in the PSNR and SSIM is higher in 1D-random under-sampling than in 2D. The requirement of incoherence for aliasing artifacts is different from that for iterative reconstruction.

LoTToR: An Algorithm for Missing-Wedge Correction of the Low-Tilt Tomographic 3D Reconstruction of a Single-Molecule Structure

A single-molecule three-dimensional (3D) structure is essential for understanding the thermal vibrations and dynamics as well as the conformational changes during the chemical reaction of macromolecules. Individual-particle electron tomography (IPET) is an approach for obtaining a snap-shot 3D structure of an individual macromolecule particle by aligning the tilt series of electron tomographic (ET) images of a targeted particle through a focused iterative 3D reconstruction method. The method can reduce the influence on the 3D reconstruction from large-scale image distortion and deformation. Due to the mechanical tilt limitation, 3D reconstruction often contains missing-wedge artifacts, presented as elongation and an anisotropic resolution. Here, we report a post-processing method to correct the missing-wedge artifact. This low-tilt tomographic reconstruction (LoTToR) method contains a model-free iteration process under a set of constraints in real and reciprocal spaces. A proof of concept is conducted by using the LoTToR on a phantom, i.e., a simulated 3D reconstruction from a low-tilt series of images, including that within a tilt range of ±15°. The method is validated by using both negative-staining (NS) and cryo-electron tomography (cryo-ET) experimental data. A significantly reduced missing-wedge artifact verifies the capability of LoTToR, suggesting a new tool to support the future study of macromolecular dynamics, fluctuation and chemical activity from the viewpoint of single-molecule 3D structure determination.

An irregular metal trace inpainting network for x-ray CT metal artifact reduction.

Metal implants in the patient's body can generate severe metal artifacts in x-ray computed tomography (CT) images. These artifacts may cover the tissues around the metal implants in CT images and even corrupt the tissue regions, thus affecting disease diagnosis using these images. Previous deep learning metal trace inpainting methods used both valid pixels of uncorrupted areas and invalid pixels of corrupted areas to patch metal trace (i.e., the holes of removed metal-corrupted regions). Such methods cannot recover fine details well and often suffer information mismatch due to interference of invalid pixels, thus incurring considerable secondary artifacts. In this paper, we develop a new irregular metal trace inpainting network for reducing metal artifacts. We develop a new deep learning network to patch irregular metal trace in metal-corrupted sinograms to reduce metal artifacts for isometric fan-beam CT. Our new method patches irregular metal trace in CT sinograms using only valid pixels, avoiding interference from invalid pixels. Furthermore, to enable the inpainting network to recover as many details as possible, we design an auxiliary inpainting network to suppress the probable secondary artifacts in CT images to assist fine detail restoration. The image produced by the auxiliary network is then projected onto a sinogram via a forward projection (FP) algorithm and is fused with the sinogram predicted by the inpainting network in order to predict the final recovered sinogram. Our entire network is trained end-to-end to extract cross-domain information between the sinogram domain and CT image domain. We compare our proposed method with two traditional and four deep learning-based metal trace inpainting methods, and with an iterative reconstruction method on four datasets: dental fillings (panoramic and local perspectives), hip prostheses, and spine fixations. We use both quantitative and qualitative indices to evaluate our method, and the analyses suggest that our method reduces the most metal artifacts and produces the best quality CT images. Additionally, our proposed method takes 0.1512s on average to process a CT slice, which meets the clinical requirement. This paper proposes a new deep learning network to patch irregular metal trace in corrupted sinograms to reduce metal artifacts. Our method restores more fine details in irregular metal trace and has a superior capability on metal artifact reduction compared with state-of-the-art methods.

Low-dose CT with deep learning regularization via proximal forward–backward splitting

Low-dose x-ray computed tomography (LDCT) is desirable for reduced patient dose. This work develops new image reconstruction methods with deep learning (DL) regularization for LDCT. Our methods are based on the unrolling of a proximal forward–backward splitting (PFBS) framework with data-driven image regularization via deep neural networks. In contrast to PFBS-IR, which utilizes standard data fidelity updates via an iterative reconstruction (IR) method, PFBS-AIR involves preconditioned data fidelity updates that fuse the analytical reconstruction (AR) and IR methods in a synergistic way, i.e. fused analytical and iterative reconstruction (AIR). The results suggest that the DL-regularized methods (PFBS-IR and PFBS-AIR) provide better reconstruction quality compared to conventional methods (AR or IR). In addition, owing to the AIR, PFBS-AIR noticeably outperformed PFBS-IR and another DL-based postprocessing method, FBPConvNet.

Differentiated Backprojection Domain Deep Learning for Conebeam Artifact Removal

Conebeam CT using a circular trajectory is quite often used for various applications due to its relative simple geometry. For conebeam geometry, Feldkamp, Davis and Kress algorithm is regarded as the standard reconstruction method, but this algorithm suffers from so-called conebeam artifacts as the cone angle increases. Various model-based iterative reconstruction methods have been developed to reduce the cone-beam artifacts, but these algorithms usually require multiple applications of computational expensive forward and backprojections. In this paper, we develop a novel deep learning approach for accurate conebeam artifact removal. In particular, our deep network, designed on the differentiated backprojection domain, performs a data-driven inversion of an ill-posed deconvolution problem associated with the Hilbert transform. The reconstruction results along the coronal and sagittal directions are then combined using a spectral blending technique to minimize the spectral leakage. Experimental results under various conditions confirmed that our method generalizes well and outperforms the existing iterative methods despite significantly reduced runtime complexity.

Limited-View and Sparse Photoacoustic Tomography for Neuroimaging with Deep Learning

Photoacoustic tomography (PAT) is a non-ionizing imaging modality capable of acquiring high contrast and resolution images of optical absorption at depths greater than traditional optical imaging techniques. Practical considerations with instrumentation and geometry limit the number of available acoustic sensors and their “view” of the imaging target, which result in image reconstruction artifacts degrading image quality. Iterative reconstruction methods can be used to reduce artifacts but are computationally expensive. In this work, we propose a novel deep learning approach termed pixel-wise deep learning (Pixel-DL) that first employs pixel-wise interpolation governed by the physics of photoacoustic wave propagation and then uses a convolution neural network to reconstruct an image. Simulated photoacoustic data from synthetic, mouse-brain, lung, and fundus vasculature phantoms were used for training and testing. Results demonstrated that Pixel-DL achieved comparable or better performance to iterative methods and consistently outperformed other CNN-based approaches for correcting artifacts. Pixel-DL is a computationally efficient approach that enables for real-time PAT rendering and improved image reconstruction quality for limited-view and sparse PAT.