Accelerated Image Reconstruction Research Articles

CMR 3-95 - Accelerated Image Reconstruction in Cardiac Cine MRI: A 3D Cnn-based Modl Approach

CMR 3-95 - Accelerated Image Reconstruction in Cardiac Cine MRI: A 3D Cnn-based Modl Approach

Designing Clinical MRI for Enhanced Workflow and Value.

MRI is an expensive and traditionally time-intensive modality in imaging. With the paradigm shift toward value-based healthcare, radiology departments must examine the entire MRI process cycle to identify opportunities to optimize efficiency and enhance value for patients. Digital tools such as "frictionless scheduling" prioritize patient preference and convenience, thereby delivering patient-centered care. Recent advances in conventional and deep learning-based accelerated image reconstruction methods have reduced image acquisition time to such a degree that so-called nongradient time now constitutes a major percentage of total room time. For this reason, architectural design strategies that reconfigure patient preparation processes and decrease the turnaround time between scans can substantially impact overall throughput while also improving patient comfort and privacy. Real-time informatics tools that provide an enterprise-wide overview of MRI workflow and Picture Archiving and Communication System (PACS)-integrated instant messaging can complement these efforts by offering transparent, situational data and facilitating communication between radiology team members. Finally, long-term investment in training, recruiting, and retaining a highly skilled technologist workforce is essential for building a pipeline and team of technologists committed to excellence. Here, we highlight various opportunities for optimizing MRI workflow and enhancing value by offering many of our own on-the-ground experiences and conclude by anticipating some of the future directions for process improvement and innovation in clinical MR imaging. EVIDENCE LEVEL: N/A TECHNICAL EFFICACY: Stage 1.

List-mode quantitative joint reconstruction of activity and attenuation maps in Time-of-Flight PET

Quantitative PET imaging requires accurate determination of patient-specific attenuation maps, which remains challenging on hybrid PET/MRI systems owing to the lack of a direct relationship between MR image intensity and attenuation coefficients. The aim of the present study is to develop a list-mode based algorithm for accurate and robust attenuation correction of PET data using time-of-flight (TOF) emission information. We analyze and address the challenges of list-mode emission-based maximum-likelihood joint estimation of activity and attenuation (LM-MLAA) in state-of-the-art PET imaging. The proposed method exploits a rapid on-the-fly system matrix calculation algorithm based on elliptic integrals while updating the attenuation map from accumulating list-mode coincidences to achieve accelerated image reconstruction. The scattering compensation is incorporated inside it using an iterative approach in such that the current estimation of attenuation map used on a course grid sampling scattering points to make an estimate of scattering. The performance of the proposed LM-MLAA approach was evaluated on Monte Carlo simulations of a phantom at different time resolutions. The contrast and noise for hot and cold regions on reconstructed images at different time resolutions were analysed. The estimated attenuation map exhibits resilience against noise, effectively eliminates high-frequency cross-talk even in the absence of prior information on attenuation coefficients, and enables discrimination among different anatomical regions in the reconstructed image. The error in the mean estimated attenuation coefficients after 50 iterations was ∼ 2% in water and ∼ -14% in Teflon regions for TOF resolutions corresponding to those of most current commercial PET systems (∼ 500 ps). The proposed LM-MLAA framework can be used for joint reconstruction of activity and attenuation maps from list-mode emission data as standalone or a complementary approach to existing in multimodality imaging such as PET/MRI, where direct measurement attenuation maps is not possible.

Accelerating image reconstruction for multi-contrast MRI based on Y-Net3.

As one of the significant preoperative imaging modalities in medical diagnosis, Magnetic resonance imaging (MRI) takes a long scanning time due to its special imaging principle. We propose an innovative MRI reconstruction strategy and data consistency method based on deep learning to reconstruct high-quality brain MRIs from down-sampled data and accelerate the MR imaging process. Sixteen healthy subjects undergoing T1-weighted spin-echo (SE) and T2-weighted fast spin-echo (FSE) sequences by a 1.5T MRI scanner were recruited. A Y-Net3+ network was used to facilitate the high-quality MRI reconstruction through context information. In addition, the existing data consistency fidelity method was improved. The difference between the reconstructed K-space and the original K-space was shorten by the linear regression algorithm. Therefore, the redundant artifacts derived from under-sampling were avoided. The Structural Similarity (SSIM) and Peak Signal to Noise Ratio (PSNR) were applied to quantitatively evaluate image reconstruction performance of different down-sampling patterns. Compared with the classical Y-Net, Y-Net3+ network improved SSIM and PSNR of MRI images from 0.9164±0.0178 and 33.2216±3.2919 to 0.9387±0.0363 and 35.1785±3.3105, respectively, under compressed sensing reconstruction with acceleration factor of 4. The improved network increases signal-to-noise ratio and adds more image texture information in the reconstructed images. Furthermore, in the process of data consistency, linear regression analysis was used to reduce the difference between the reconstructed K-space and the original K-space, so that the SSIM and PSNR were increased to 0.9808±0.0081 and 40.9254±1.1911, respectively. The improved Y-Net combined with data consistency fidelity method elucidates its potential in reconstructing high-quality T2-weighted images from the down-sampled data by fully exploring the T1-weighted information. With the advantage of avoiding down-sampled artifacts, the improved network exhibits remarkable clinical promise for fast MRI applications.

Deep Learning Accelerated Image Reconstruction of Fluid-Attenuated Inversion Recovery Sequence in Brain Imaging: Reduction of Acquisition Time and Improvement of Image Quality

Fluid-attenuated inversion recovery (FLAIR) imaging is playing an increasingly significant role in the detection of brain metastases with a concomitant increase in the number of magnetic resonance imaging (MRI) examinations. Therefore, the purpose of this study was to investigate the impact on image quality and diagnostic confidence of an innovative deep learning-based accelerated FLAIR (FLAIRDLR) sequence of the brain compared to conventional (standard) FLAIR (FLAIRS) imaging. Seventy consecutive patients with staging cerebral MRIs were retrospectively enrolled in this single-center study. The FLAIRDLR was conducted using the same MRI acquisition parameters as the FLAIRS sequence, except for a higher acceleration factor for parallel imaging (from 2 to 4), which resulted in a shorter acquisition time of 1:39minute instead of 2:40minutes (-38%). Two specialized neuroradiologists evaluated the imaging datasets using a Likert scale that ranged from 1 to 4, with 4 indicating the best score for the following parameters: sharpness, lesion demarcation, artifacts, overall image quality, and diagnostic confidence. Additionally, the image preference of the readers and the interreader agreement were assessed. The average age of the patients was 63±11years. FLAIRDLR exhibited significantly less image noise than FLAIRS, with P-values of<.001 and<.05, respectively. The sharpness of the images and the ability to detect lesions were rated higher in FLAIRDLR, with a median score of 4 compared to a median score of 3 in FLAIRS (P-values of<.001 for both readers). In terms of overall image quality, FLAIRDLR was rated superior to FLAIRS, with a median score of 4 vs 3 (P-values of<.001 for both readers). Both readers preferred FLAIRDLR in 68/70 cases. The feasibility of deep learning FLAIR brain imaging was shown with additional 38% reduction in examination time compared to standard FLAIR imaging. Furthermore, this technique has shown improvement in image quality, noise reduction, and lesion demarcation.

Energy-efficient high-fidelity image reconstruction with memristor arrays for medical diagnosis

Medical imaging is an important tool for accurate medical diagnosis, while state-of-the-art image reconstruction algorithms raise critical challenges in massive data processing for high-speed and high-quality imaging. Here, we present a memristive image reconstructor (MIR) to greatly accelerate image reconstruction with discrete Fourier transformation (DFT) by computing-in-memory (CIM) with memristor arrays. A high-accuracy quasi-analogue mapping (QAM) method and generic complex matrix transfer (CMT) scheme was proposed to improve the mapping precision and transfer efficiency, respectively. High-fidelity magnetic resonance imaging (MRI) and computed tomography (CT) image reconstructions were demonstrated, achieving software-equivalent qualities and DICE scores after segmentation with nnU-Net algorithm. Remarkably, our MIR exhibited 153× and 79× improvements in energy efficiency and normalized image reconstruction speed, respectively, compared to graphics processing unit (GPU). This work demonstrates MIR as a promising high-fidelity image reconstruction platform for future medical diagnosis, and also largely extends the application of memristor-based CIM beyond artificial neural networks.

Modified Polar Grid-based Accelerated Image Reconstruction Technique for X-ray CT

Image reconstruction by projections is basically a mathematical problem. The speed and accuracy of the reconstruction mainly depend upon, (1) discretization scheme (2) solution technique. Solution by Algebraic Method (AM) is the most promising technique for manufacturing the low-cost (few detectors arrangement) X-ray CT set-up. Slow converging rate and huge memory requirement are the two major drawbacks of the AM algorithms. Projection coefficient calculations and their storages are major time and storage-consuming processes of the algebraic methods. Among all algebraic methods, the Multiplicative Algebraic Reconstruction Technique (MART) is more effective because it maximizes the entropy of the image space. The conventional MART algorithm uses square grids (SG) in the discretization process of the image reconstruction. In the present work, we employed a modified polar grid to cope with the shortcomings of the MART method. In this new discretization scheme, only the projection coefficients of the first view, need to be calculated. Projection coefficients for other views can be easily calculated by using the symmetries of the modified polar grid. This method significantly reduces the reconstruction time and storage requirement of the MART algorithm. In addition, we also presented the direct method of polar to square grid transformation for the visualization of the reconstructed images. We tested the proposed method to fan-beam geometry for 2D image reconstruction. Structural similarity index(SSIM) and L2 error are used for quality assessment of the reconstructed images.

An LED-Based structured illumination microscope using a digital micromirror device and GPU accelerated image reconstruction.

When combined with computational approaches, fluorescence imaging becomes one of the most powerful tools in biomedical research. It is possible to achieve resolution figures beyond the diffraction limit, and improve the performance and flexibility of high-resolution imaging systems with techniques such as structured illumination microscopy (SIM) reconstruction. In this study, the hardware and software implementation of an LED-based super-resolution imaging system using SIM employing GPU accelerated parallel image reconstruction is presented. The sample is illuminated with two-dimensional sinusoidal patterns with various orientations and lateral phase shifts generated using a digital micromirror device (DMD). SIM reconstruction is carried out in frequency space using parallel CUDA kernel functions. Furthermore, a general purpose toolbox for the parallel image reconstruction algorithm and an infrastructure that allows all users to perform parallel operations on images without developing any CUDA kernel code is presented. The developed image reconstruction algorithm was run separately on a CPU and a GPU. Two different SIM reconstruction algorithms have been developed for the CPU as mono-thread CPU algorithm and multi-thread OpenMP CPU algorithm. SIM reconstruction of 1024 × 1024 px images was achieved in 1.49 s using GPU computation, indicating an enhancement by ∼28 and ∼20 in computation time when compared with mono-thread CPU computation and multi-thread OpenMP CPU computation, respectively.

Photon-Counting CT Reconstruction With a Learned Forward Operator

Photon-Counting CT is an emerging imaging technology that promises higher spatial resolution and the possibility for material decomposition in the reconstruction. A major difficulty in Photon-Counting CT is to efficiently model cross-talk between detectors. In this work, we accelerate image reconstruction tasks for Photon-Counting CT by modelling the cross-talk with an appropriately trained deep convolutional neural network. The main result relates to proving convergence when using such a learned cross-talk model in the context of second-order optimisation methods for spectral CT. Another is to evaluate the method through numerical experiments on small-scale CT acquisitions generated using a realistic physics model. Using the reconstruction with a full cross-talk model as ground truth, the learned cross-talk model results in a 20&#x00A0;dB increase in peak-signal-to noise ratio compared to ignoring cross-talk altogether. At the same time, it effectively cuts the computation time of the full cross-talk model in half. Furthermore, the learned cross-talk model generalises well to both unseen data and unseen detector settings. Our results indicate that such a partially learned forward operator is a suitable way of modelling data generation in Photon-Counting CT with a computational benefit that becomes more noticeable for realistic problem sizes.

Accelerated image reconstruction with separable Hankel regularization in parallel MRI.

Magnetic resonance imaging has been widely adopted in clinical diagnose, however, it suffers from relatively long data acquisition time. Sparse sampling with reconstruction can speed up the data acquisition duration. As the state-of-the-art magnetic resonance imaging methods, the structured low rank reconstruction approaches embrace the advantage of holding low reconstruction errors and permit flexible undersampling patterns. However, this type of method demands intensive computations and high memory consumptions, thereby resulting in a lengthy reconstruction time. In this work, we proposed a separable Hankel low rank reconstruction method to explore the low rankness of each row and each column. Furthermore, we utilized the self-consistence and conjugate symmetry property of k-space data. The experimental results demonstrated that the proposed method outperforms the state-of-the-art approaches in terms of lower reconstruction errors and better detail preservation. Besides, the proposed method requires much less computation and memory consumption.Clinical Relevance- Parallel imaging, image reconstruction, Hankel low-rank.

CORE-Deblur: Parallel MRI Reconstruction by Deblurring using compressed sensing

In this work we introduce a new method that combines Parallel MRI and Compressed Sensing (CS) for accelerated image reconstruction from subsampled k-space data. The method first computes a convolved image, which gives the convolution between a user-defined kernel and the unknown MR image, and then reconstructs the image by CS-based image deblurring, in which CS is applied for removing the inherent blur stemming from the convolution process. This method is hence termed CORE-Deblur. Retrospective subsampling experiments with data from a numerical brain phantom and in-vivo 7T brain scans showed that CORE-Deblur produced high-quality reconstructions, comparable to those of a conventional CS method, while reducing the number of iterations by a factor of 10 or more. The average Normalized Root Mean Square Error (NRMSE) obtained by CORE-Deblur for the in-vivo datasets was 0.016. CORE-Deblur also exhibited robustness regarding the chosen kernel and compatibility with various k-space subsampling schemes, ranging from regular to random. In summary, CORE-Deblur enables high quality reconstructions and reduction of the CS iterations number by 10-fold.

Accelerated image reconstruction by a combined dual-matrix dual-voxel approach

Today, versatile emission computed tomography (VECTor) technology using dedicated high-energy collimation enables simultaneous positron emission tomography (PET) and single photon emission computed tomography (SPECT) down to 0.6 mm and 0.4 mm resolution in mice, respectively. We recently showed that for optimal resolution and quantitative accuracy of PET images the long tails of the 511 keV point spread functions (PSFs) need to be fully modelled during image reconstruction. This, however, leads to very time consuming reconstructions and thus significant acceleration in reconstruction speed is highly desirable. To this end we propose and validate a combined dual-matrix dual-voxel (DM-DV) approach: for the forward projection the slowly varying PSF tails are modelled on a three times rougher voxel grid than the central parts of the PSFs, while in the backprojection only parts of the PSF tails are included. DM-DV reconstruction is implemented in pixel-based ordered subsets expectation maximization (POSEM) and in a recently proposed accelerated pixel-based similarity-regulated ordered subsets expectation maximization (SROSEM). Both a visual assessment and a quantitative contrast-noise analysis confirm that images of a hot-rod phantom are practically identical when reconstructed with standard POSEM, DM-DV-POSEM or DM-DV-SROSEM. However, compared to POSEM, DM-DV-POSEM can reach the same contrast 5.0 times faster, while with DM-DV-SROSEM this acceleration factor increases to 11.5. Furthermore, mouse cardiac and bone images reconstructed with DM-DV-SROSEM are visually almost indistinguishable from POSEM reconstructed images but typically need an order of magnitude less reconstruction time.

First-order approximation of fluorescence excitation-transmission to accelerate fluorescence molecular tomography image reconstruction.

We present a method to accelerate image reconstruction in fluorescence molecular tomography based on the historical path fluorescence Monte Carlo model. The method exploits a first-order approximation expression during the fluorescence excitation-transmission process to merge the path and state information of the photon in a voxel. The experiments show that our method not only greatly reduces the amount of data required for storage in the hard disk and accelerates image reconstruction, but also maintains the quantitative and positioning accuracy of the conventional method.

Accelerated image reconstruction using extrapolated Tikhonov filtering for photoacoustic tomography.

Development of simple and computationally efficient extrapolated Tikhonov filtering reconstruction methods for photoacoustic tomography. The model-based reconstruction algorithms in photoacoustic tomography typically utilize Tikhonov regularization scheme for the reconstruction of initial pressure distribution from the measured boundary acoustic data. The automated choice of regularization parameter in these cases is computationally expensive. Moreover, the Tikhonov scheme promotes the smooth features in the reconstructed image due to the smooth regularizer, thus leading to loss of sharp features. The proposed extrapolation method estimates the solution at zero regularization assuming the solution being a function of regularization parameter and thus posing it as a zero value problem. Thus, the numerically computed zero regularization solution is expected to have better features compared to standard Tikhonov solution, with an added advantage of removing the necessity of automated choice of regularization. The reconstructed results using this method were shown in three variants (Lanczos, traditional, and exponential) of Tikhonov filtering and were compared with the standard error estimate technique. Four numerical (including realistic breast phantom) and two experimental phantom data were utilized to show the effectiveness of the proposed method. It was shown that the proposed method performance was superior than the standard error estimate technique, being at least four times faster in terms of computation, and provides an improvement as high as 2.6 times in terms of standard figures of merit. The developed extrapolated Tikhonov filtering methods overcome the difficulty of obtaining a suitable regularization parameter and shown to be reconstructing high-quality photoacoustic images with additional advantage of being computationally efficient, making it more appealing in real-time applications.

Abstracts of the Total Body PET conference 2018

Abstracts of the Total Body PET conference 2018

Accelerated reconstruction of electrical impedance tomography images via patch based sparse representation.

Electrical impedance tomography (EIT) reconstruction is a nonlinear and ill-posed problem. Exact reconstruction of an EIT image inverts a high dimensional mathematical model to calculate the conductivity field, which causes significant problems regarding that the computational complexity will reduce the achievable frame rate, which is considered as a major advantage of EIT imaging. The single-step method, state estimation method, and projection method were always used to accelerate reconstruction process. The basic principle of these methods is to reduce computational complexity. However, maintaining high resolution in space together with not much cost is still challenging, especially for complex conductivity distribution. This study proposes an idea to accelerate image reconstruction of EIT based on compressive sensing (CS) theory, namely, CSEIT method. The novel CSEIT method reduces the sampling rate through minimizing redundancy in measurements, so that detailed information of reconstruction is not lost. In order to obtain sparse solution, which is the prior condition of signal recovery required by CS theory, a novel image reconstruction algorithm based on patch-based sparse representation is proposed. By applying the new framework of CSEIT, the data acquisition time, or the sampling rate, is reduced by more than two times, while the accuracy of reconstruction is significantly improved.

Similarity-regulation of OS-EM for accelerated SPECT reconstruction

Ordered subsets expectation maximization (OS-EM) is widely used to accelerate image reconstruction in single photon emission computed tomography (SPECT). Speedup of OS-EM over maximum likelihood expectation maximization (ML-EM) is close to the number of subsets used. Although a high number of subsets can shorten reconstruction times significantly, it can also cause severe image artifacts such as improper erasure of reconstructed activity if projections contain few counts. We recently showed that such artifacts can be prevented by using a count-regulated OS-EM (CR-OS-EM) algorithm which automatically adapts the number of subsets for each voxel based on the estimated number of counts that the voxel contributed to the projections. While CR-OS-EM reached high speed-up over ML-EM in high-activity regions of images, speed in low-activity regions could still be very slow. In this work we propose similarity-regulated OS-EM (SR-OS-EM) as a much faster alternative to CR-OS-EM. SR-OS-EM also automatically and locally adapts the number of subsets, but it uses a different criterion for subset regulation: the number of subsets that is used for updating an individual voxel depends on how similar the reconstruction algorithm would update the estimated activity in that voxel with different subsets. Reconstructions of an image quality phantom and in vivo scans show that SR-OS-EM retains all of the favorable properties of CR-OS-EM, while reconstruction speed can be up to an order of magnitude higher in low-activity regions. Moreover our results suggest that SR-OS-EM can be operated with identical reconstruction parameters (including the number of iterations) for a wide range of count levels, which can be an additional advantage from a user perspective since users would only have to post-filter an image to present it at an appropriate noise level.

Accelerated image reconstruction in fluorescence molecular tomography using a nonuniform updating scheme with momentum and ordered subsets methods.

Fluorescence molecular tomography (FMT) is a significant preclinical imaging modality that has been actively studied in the past two decades. It remains a challenging task to obtain fast and accurate reconstruction of fluorescent probe distribution in small animals due to the large computational burden and the ill-posed nature of the inverse problem. We have recently studied a nonuniform multiplicative updating algorithm that combines with the ordered subsets (OS) method for fast convergence. However, increasing the number of OS leads to greater approximation errors and the speed gain from larger number of OS is limited. We propose to further enhance the convergence speed by incorporating a first-order momentum method that uses previous iterations to achieve optimal convergence rate. Using numerical simulations and a cubic phantom experiment, we have systematically compared the effects of the momentum technique, the OS method, and the nonuniform updating scheme in accelerating the FMT reconstruction. We found that the proposed combined method can produce a high-quality image using an order of magnitude less time.

Implementation of GPU-accelerated back projection for EPR imaging.

Electron paramagnetic resonance (EPR) Imaging (EPRI) is a robust method for measuring in vivo oxygen concentration (pO2). For 3D pulse EPRI, a commonly used reconstruction algorithm is the filtered backprojection (FBP) algorithm, in which the backprojection process is computationally intensive and may be time consuming when implemented on a CPU. A multistage implementation of the backprojection can be used for acceleration, however it is not flexible (requires equal linear angle projection distribution) and may still be time consuming. In this work, single-stage backprojection is implemented on a GPU (Graphics Processing Units) having 1152 cores to accelerate the process. The GPU implementation results in acceleration by over a factor of 200 overall and by over a factor of 3500 if only the computing time is considered. Some important experiences regarding the implementation of GPU-accelerated backprojection for EPRI are summarized. The resulting accelerated image reconstruction is useful for real-time image reconstruction monitoring and other time sensitive applications.

GPU Accelerated Image Reconstruction in a Two-Strip J-PET Tomograph

We present a fast GPU implementation of the image reconstruction routine, for a novel two strip PET detector that relies solely on the time of flight measurements.