Parallel Imaging Reconstruction Research Articles

GPU-accelerated parallel image reconstruction strategies for magnetic particle imaging

Objective. Image reconstruction is a fundamental step in magnetic particle imaging (MPI). One of the main challenges is the fact that the reconstructions are computationally intensive and time-consuming, so choosing an algorithm presents a compromise between accuracy and execution time, which depends on the application. This work proposes a method that provides both fast and accurate image reconstructions. Approach. Image reconstruction algorithms were implemented to be executed in parallel in graphics processing units (GPUs) using the CUDA framework. The calculation of the model-based MPI calibration matrix was also implemented in GPU to allow both fast and flexible reconstructions. Main results. The parallel algorithms were able to accelerate the reconstructions by up to about 6,100 times in comparison to the serial Kaczmarz algorithm executed in the CPU, allowing for real-time applications. Reconstructions using the OpenMPIData dataset validated the proposed algorithms and demonstrated that they are able to provide both fast and accurate reconstructions. The calculation of the calibration matrix was accelerated by up to about 37 times. Significance. The parallel algorithms proposed in this work can provide single-frame MPI reconstructions in real time, with frame rates greater than 100 frames per second. The parallel calculation of the calibration matrix can be combined with the parallel reconstruction to deliver images in less time than the serial Kaczmarz reconstruction, potentially eliminating the need of storing the calibration matrix in the main memory, and providing the flexibility of redefining scanning and reconstruction parameters during execution.

DFUSNN: zero-shot dual-domain fusion unsupervised neural network for parallel MRI reconstruction

Objective. Recently, deep learning models have been used to reconstruct parallel magnetic resonance (MR) images from undersampled k-space data. However, most existing approaches depend on large databases of fully sampled MR data for training, which can be challenging or sometimes infeasible to acquire in certain scenarios. The goal is to develop an effective alternative for improved reconstruction quality that does not rely on external training datasets. Approach. We introduce a novel zero-shot dual-domain fusion unsupervised neural network (DFUSNN) for parallel MR imaging reconstruction without any external training datasets. We employ the Noise2Noise (N2N) network for the reconstruction in the k-space domain, integrate phase and coil sensitivity smoothness priors into the k-space N2N network, and use an early stopping criterion to prevent overfitting. Additionally, we propose a dual-domain fusion method based on Bayesian optimization to enhance reconstruction quality efficiently. Results. Simulation experiments conducted on three datasets with different undersampling patterns showed that the DFUSNN outperforms all other competing unsupervised methods and the one-shot Hankel-k-space generative model (HKGM). The DFUSNN also achieves comparable results to the supervised Deep-SLR method. Significance. The novel DFUSNN model offers a viable solution for reconstructing high-quality MR images without the need for external training datasets, thereby overcoming a major hurdle in scenarios where acquiring fully sampled MR data is difficult.

In-line imaging and recognition of flip chip fabrication defects by real-time photoacoustic remote sensing system

Microscopic defects in flip chips, originating from manufacturing, significantly affect performance and longevity. Post-fabrication sampling methods ensure product functionality but lack in-line defect monitoring to enhance chip yield and lifespan in real-time. This study introduces a photoacoustic remote sensing (PARS) system for in-line imaging and defect recognition during flip-chip fabrication. We first propose a real-time PARS imaging method based on continuous acquisition combined with parallel processing image reconstruction to achieve real-time imaging during the scanning of flip-chip samples, reducing reconstruction time from an average of approximately 1134 ms to 38 ms. Subsequently, we propose improved YOLOv7 with space-to-depth block (IYOLOv7-SPD), an enhanced deep learning defect recognition method, for accurate in-line recognition and localization of microscopic defects during the PARS real-time imaging process. The experimental results validate the viability of the proposed system for enhancing the lifespan and yield of flip-chip products in chip manufacturing facilities.

Deep learning image reconstruction of diffusion-weighted imaging in evaluation of prostate cancer focusing on its clinical implications.

Image-based assessment of prostate cancer (PCa) is increasingly emphasized in the diagnostic workflow for selecting biopsy targets and possibly predicting clinically significant prostate cancer (csPCa). Assessment is based on Prostate Imaging-Reporting and Data System (PI-RADS) which is largely dependent on T2-weighted image (T2WI) and diffusion weighted image (DWI). This study aims to determine whether deep learning reconstruction (DLR) can improve the image quality of DWI and affect the assessment of PI-RADS ≥4 in patients with PCa. In this retrospective study, 3.0T post-biopsy prostate magnetic resonance imaging (MRI) of 70 patients with PCa in Korea University Ansan Hospital from November 2021 to July 2022 was reconstructed with and without using DLR. Four DWI image sets were made: (I) conventional DWI (CDWI): DWI with acceleration factor 2 and conventional parallel imaging reconstruction, (II) DL1: DWI with acceleration factor 2 using DLR, (III) DL2: DWI with acceleration factor 3 using DLR, and (IV) DL3: DWI with acceleration factor 3 and half average b-value using DLR. Apparent diffusion coefficient (ADC) value, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were measured by one reviewer, while two reviewers independently assessed overall image quality, noise, and lesion conspicuity using a four-point visual scoring system from each DWI image set. Two reviewers also performed PI-RADSv2.1 scoring on lesions suspected of malignancy. A total of 70 patients (mean age, 70.8±9.7 years) were analyzed. The image acquisition time was 4:46 min for CDWI and DL1, 3:40 min for DL2, and 2:00 min for DL3. DL1 and DL2 images resulted in better lesion conspicuity compared to CDWI images assessed by both readers (P<0.05). DLR resulted in a significant increase in SNR, from 38.4±14.7 in CDWI to 56.9±21.0 in DL1. CNR increased from 25.1±11.5 in CDWI to 43.1±17.8 in DL1 (P<0.001). PI-RADS v2.1 scoring for PCa lesions was more agreeable with the DL1 reconstruction method than with CDWI (κ value CDWI, DL1; 0.40, 0.61, respectively). A statistically significant number of lesions were upgraded from PI-RADS <4 in CDWI image to PI-RADS ≥4 in DL1 images for both readers (P<0.05). Most of the PI-RADS upgraded lesions were from higher than unfavorable intermediate-risk groups according to the 2023 National Comprehensive Cancer Network guidelines with statistically significant difference of marginal probability in DL1 and DL2 for both readers (P<0.05). DLR in DWI for PCa can provide options for improving image quality with a significant impact on PI-RADS evaluation or about a 23% reduction in acquisition time without compromising image quality.

Improved Complex Convolutional Neural Network Based on SPIRiT and Dense Connection for Parallel MRI Reconstruction

To accelerate the data acquisition speed of magnetic resonance imaging (MRI) and improve the reconstructed MR images’ quality, we propose a parallel MRI reconstruction model (SPIRiT-Net), which combines the iterative self-consistent parallel imaging reconstruction model (SPIRiT) with the cascaded complex convolutional neural networks (CCNNs). More specifically, this model adopts the SPIRiT model for reconstruction in the k-space domain and the cascaded CCNNs with dense connection for reconstruction in the image domain. Meanwhile, this model introduces the data consistency layers for better reconstruction in both the image domain and the k-space domain. The experimental results on two clinical knee datasets as well as the fastMRI brain dataset under different undersampling patterns show that the SPIRiT-Net model achieves better reconstruction performance in terms of visual effect, peak signal-to-noise ratio, and structural similarity over SPIRiT, Deepcomplex, and DONet. It will be beneficial to the diagnosis of clinical medicine.

Dynamic parallel imaging at 9.4 T using reconfigurable receive coaxial dipoles.

Parallel imaging is one of the key MRI technologies that allow reduction of image acquisition time. However, the parallel imaging reconstruction commonly leads to a signal-to-noise ratio (SNR) drop evaluated using a so-called geometrical factor (g-factor). The g-factor is minimized by increasing the number of array elements and their spatial diversity. At the same time, increasing the element count requires a decrease in their size. This may lead to insufficient coil loading, an increase in the relative noise contribution from the RF coil itself, and hence SNR reduction. Previously, instead of increasing the channel number, we introduced the concept of electronically switchable time-varying sensitivities, which was shown to improve parallel imaging performance. In this approach, each reconfigurable receive element supports two spatially distinct sensitivity profiles. In this work, we developed and evaluated a novel eight-element human head receive-only reconfigurable coaxial dipole array for human head imaging at 9.4 T. In contrast to the previously reported reconfigurable dipole array, the new design does not include direct current (DC) control wires connected directly to the dipoles. The coaxial cable itself is used to deliver DC voltage to the PIN diodes located at the ends of the antennas. Thus, the novel reconfigurable coaxial dipole design opens a way to scale the dynamic parallel imaging up to a realistic number of channels, that is, 32 and above. The novel array was optimized and tested experimentally, including in vivo studies. It was found that dynamic sensitivity switching provided an 8% lower mean and 33% lower maximum g-factor (for Ry × Rz = 2 × 2 acceleration) compared with conventional static sensitivities.

DONet: Dual-Octave Network for Fast MR Image Reconstruction.

Magnetic resonance (MR) image acquisition is an inherently prolonged process, whose acceleration has long been the subject of research. This is commonly achieved by obtaining multiple undersampled images, simultaneously, through parallel imaging. In this article, we propose the dual-octave network (DONet), which is capable of learning multiscale spatial-frequency features from both the real and imaginary components of MR data, for parallel fast MR image reconstruction. More specifically, our DONet consists of a series of dual-octave convolutions (Dual-OctConvs), which are connected in a dense manner for better reuse of features. In each Dual-OctConv, the input feature maps and convolutional kernels are first split into two components (i.e., real and imaginary) and then divided into four groups according to their spatial frequencies. Then, our Dual-OctConv conducts intragroup information updating and intergroup information exchange to aggregate the contextual information across different groups. Our framework provides three appealing benefits: 1) it encourages information interaction and fusion between the real and imaginary components at various spatial frequencies to achieve richer representational capacity; 2) the dense connections between the real and imaginary groups in each Dual-OctConv make the propagation of features more efficient by feature reuse; and 3) DONet enlarges the receptive field by learning multiple spatial-frequency features of both the real and imaginary components. Extensive experiments on two popular datasets (i.e., clinical knee and fastMRI), under different undersampling patterns and acceleration factors, demonstrate the superiority of our model in accelerated parallel MR image reconstruction.

Fast reconstruction of SMS bSSFP myocardial perfusion images using noise map estimation network (NoiseMapNet): a head-to-head comparison with parallel imaging and iterative reconstruction.

Simultaneous multi-slice (SMS) bSSFP imaging enables stress myocardial perfusion imaging with high spatial resolution and increased spatial coverage. Standard parallel imaging techniques (e.g., TGRAPPA) can be used for image reconstruction but result in high noise level. Alternatively, iterative reconstruction techniques based on temporal regularization (ITER) improve image quality but are associated with reduced temporal signal fidelity and long computation time limiting their online use. The aim is to develop an image reconstruction technique for SMS-bSSFP myocardial perfusion imaging combining parallel imaging and image-based denoising using a novel noise map estimation network (NoiseMapNet), which preserves both sharpness and temporal signal profiles and that has low computational cost. The proposed reconstruction of SMS images consists of a standard temporal parallel imaging reconstruction (TGRAPPA) with motion correction (MOCO) followed by image denoising using NoiseMapNet. NoiseMapNet is a deep learning network based on a 2D Unet architecture and aims to predict a noise map from an input noisy image, which is then subtracted from the noisy image to generate the denoised image. This approach was evaluated in 17 patients who underwent stress perfusion imaging using a SMS-bSSFP sequence. Images were reconstructed with (a) TGRAPPA with MOCO (thereafter referred to as TGRAPPA), (b) iterative reconstruction with integrated motion compensation (ITER), and (c) proposed NoiseMapNet-based reconstruction. Normalized mean squared error (NMSE) with respect to TGRAPPA, myocardial sharpness, image quality, perceived SNR (pSNR), and number of diagnostic segments were evaluated. NMSE of NoiseMapNet was lower than using ITER for both myocardium (0.045 ± 0.021 vs. 0.172 ± 0.041, p < 0.001) and left ventricular blood pool (0.025 ± 0.014 vs. 0.069 ± 0.020, p < 0.001). There were no significant differences between all methods for myocardial sharpness (p = 0.77) and number of diagnostic segments (p = 0.36). ITER led to higher image quality than NoiseMapNet/TGRAPPA (2.7 ± 0.4 vs. 1.8 ± 0.4/1.3 ± 0.6, p < 0.001) and higher pSNR than NoiseMapNet/TGRAPPA (3.0 ± 0.0 vs. 2.0 ± 0.0/1.3 ± 0.6, p < 0.001). Importantly, NoiseMapNet yielded higher pSNR (p < 0.001) and image quality (p < 0.008) than TGRAPPA. Computation time of NoiseMapNet was only 20s for one entire dataset. NoiseMapNet-based reconstruction enables fast SMS image reconstruction for stress myocardial perfusion imaging while preserving sharpness and temporal signal profiles.

One-Shot Generative Prior in Hankel-k-Space for Parallel Imaging Reconstruction.

Magnetic resonance imaging serves as an essential tool for clinical diagnosis. However, it suffers from a long acquisition time. The utilization of deep learning, especially the deep generative models, offers aggressive acceleration and better reconstruction in magnetic resonance imaging. Nevertheless, learning the data distribution as prior knowledge and reconstructing the image from limited data remains challenging. In this work, we propose a novel Hankel-k-space generative model (HKGM), which can generate samples from a training set of as little as one k-space. At the prior learning stage, we first construct a large Hankel matrix from k-space data, then extract multiple structured k-space patches from the Hankel matrix to capture the internal distribution among different patches. Extracting patches from a Hankel matrix enables the generative model to be learned from the redundant and low-rank data space. At the iterative reconstruction stage, the desired solution obeys the learned prior knowledge. The intermediate reconstruction solution is updated by taking it as the input of the generative model. The updated result is then alternatively operated by imposing low-rank penalty on its Hankel matrix and data consistency constraint on the measurement data. Experimental results confirmed that the internal statistics of patches within single k-space data carry enough information for learning a powerful generative model and providing state-of-the-art reconstruction.

WKGM: weighted k-space generative model for parallel imaging reconstruction.

Deep learning based parallel imaging (PI) has made great progress in recent years to accelerate MRI. Nevertheless, it still has some limitations: for example, the robustness and flexibility of existing methods are greatly deficient. In this work, we propose a method to explore the k-space domain learning via robust generative modeling for flexible calibrationless PI reconstruction, coined the weighted k-space generative model (WKGM). Specifically, WKGM is a generalized k-space domain model, where the k-space weighting technology and high-dimensional space augmentation design are efficiently incorporated for score-based generative model training, resulting in good and robust reconstructions. In addition, WKGM is flexible and thus can be synergistically combined with various traditional k-space PI models, which can make full use of the correlation between multi-coil data and realize calibrationless PI. Even though our model was trained on only 500 images, experimental results with varying sampling patterns and acceleration factors demonstrate that WKGM can attain state-of-the-art reconstruction results with the well learned k-space generative prior.

Bayesian reconstruction of magnetic resonance images using Gaussian processes

A central goal of modern magnetic resonance imaging (MRI) is to reduce the time required to produce high-quality images. Efforts have included hardware and software innovations such as parallel imaging, compressed sensing, and deep learning-based reconstruction. Here, we propose and demonstrate a Bayesian method to build statistical libraries of magnetic resonance (MR) images in k-space and use these libraries to identify optimal subsampling paths and reconstruction processes. Specifically, we compute a multivariate normal distribution based upon Gaussian processes using a publicly available library of T1-weighted images of healthy brains. We combine this library with physics-informed envelope functions to only retain meaningful correlations in k-space. This covariance function is then used to select a series of ring-shaped subsampling paths using Bayesian optimization such that they optimally explore space while remaining practically realizable in commercial MRI systems. Combining optimized subsampling paths found for a range of images, we compute a generalized sampling path that, when used for novel images, produces superlative structural similarity and error in comparison to previously reported reconstruction processes (i.e. 96.3% structural similarity and < 0.003 normalized mean squared error from sampling only 12.5% of the k-space data). Finally, we use this reconstruction process on pathological data without retraining to show that reconstructed images are clinically useful for stroke identification. Since the model trained on images of healthy brains could be directly used for predictions in pathological brains without retraining, it shows the inherent transferability of this approach and opens doors to its widespread use.

Low-rank tensor assisted K-space generative model for parallel imaging reconstruction

Although recent deep learning methods, especially generative models, have shown good performance in magnetic resonance imaging, there is still much room for improvement. Considering that the sample number and internal dimension in score-based generative model have key influence on estimating the gradients of data distribution, we present a new idea for parallel imaging reconstruction, named low-rank tensor assisted k-space generative model (LR-KGM). It means that we transform low-rank information into high-dimensional prior information for learning. More specifically, the multi-channel data is constructed into a large Hankel matrix to reduce the number of training samples, which is subsequently collapsed into a tensor for the stage of prior learning. In the testing phase, the low-rank rotation strategy is utilized to impose low-rank constraints on the output tensors of the generative network. Furthermore, we alternate the reconstruction between traditional generative iterations and low-rank high-dimensional tensor iterations. Experimental comparisons with the state-of-the-arts demonstrated that the proposed LR-KGM method achieved better performance.

Improved Simultaneous Multi-slice imaging with Composition of k-space Interpolations (SMS-COOKIE) for myocardial T1 mapping.

The aim of this study is to develop and evaluate a regularized Simultaneous Multi-Slice (SMS) reconstruction method for improved Cardiac Magnetic Resonance Imaging (CMR). The proposed reconstruction method, SMS with COmpOsition of k-space IntErpolations (SMS-COOKIE) combines the advantages of Iterative Self-consistent Parallel Imaging Reconstruction (SPIRiT) and split slice-Generalized Autocalibrating Partially Parallel Acquisitions (GRAPPA), while allowing regularization for further noise reduction. The proposed SMS-COOKIE was implemented with and without regularization, and validated using a Saturation Pulse-Prepared Heart rate Independent inversion REcovery (SAPPHIRE) myocardial T1 mapping sequence. The performance of the proposed reconstruction method was compared to ReadOut (RO)-SENSE-GRAPPA and split slice-GRAPPA, on both retrospectively and prospectively three-fold SMS-accelerated data with an additional two-fold in-plane acceleration. All SMS reconstruction methods yielded similar T1 values compared to single band imaging. SMS-COOKIE showed lower spatial variability in myocardial T1 with significant improvement over RO-SENSE-GRAPPA and split slice-GRAPPA (P < 10-4). The proposed method with additional locally low rank (LLR) regularization reduced the spatial variability, again with significant improvement over RO-SENSE-GRAPPA and split slice-GRAPPA (P < 10-4). In conclusion, improved reconstruction quality was achieved with the proposed SMS-COOKIE, which also provided lower spatial variability with significant improvement over split slice-GRAPPA.

Fast and Calibrationless Low-Rank Parallel Imaging Reconstruction through Unrolled Deep Learning Estimation of Multi-Channel Spatial Support Maps.

Low-rank technique has emerged as a powerful calibrationless alternative for parallel magnetic resonance (MR) imaging. Calibrationless low-rank reconstruction, such as low-rank modeling of local k-space neighborhoods (LORAKS), implicitly exploits both coil sensitivity modulations and the finite spatial support constraint of MR images through an iterative low-rank matrix recovery process. Although powerful, this slow iteration process is computationally demanding and reconstruction requires empirical rank optimization, hampering its robust applications for high-resolution volume imaging. This paper proposes a fast and calibrationless low-rank reconstruction of undersampled multi-slice MR brain data, based on the finite spatial support constraint reformulation with a direct deep learning estimation of spatial support maps. The iteration process of low-rank reconstruction is unrolled into a complex-valued network by training on fully-sampled multi-slice axial brain datasets acquired from the same MR coil system. To utilize coil-subject geometric parameters available for datasets, the model minimizes a hybrid loss on two sets of spatial support maps, corresponding to brain data at the original slice locations as actually acquired and nearby locations within the standard reference coordinate. This deep learning framework was integrated with LORAKS reconstruction and was evaluated with publically available gradient-echo T1-weighted brain datasets. It directly produced high-quality multi-channel spatial support maps from undersampled data, enabling rapid reconstruction without iteration. Moreover, it led to effective reductions of artifacts and noise amplification at high acceleration. In summary, our proposed deep learning framework offers a new strategy to advance the existing calibrationless low-rank reconstruction, rendering it computationally efficient, simple, and robust in practice.

Parallel imaging reconstruction using spatial nulling maps.

To develop a robust parallel imaging reconstruction method using spatial nulling maps (SNMs). Parallel reconstruction using null operations (PRUNO) is a k-space reconstruction method where a k-space nulling system is derived using null-subspace bases of the calibration matrix. ESPIRiT reconstruction extends the PRUNO subspace concept by exploiting the linear relationship between signal-subspace bases and spatial coil sensitivity characteristics, yielding a hybrid-domain approach. Yet it requires empirical eigenvalue thresholding to mask the coil sensitivity information and is sensitive to signal- and null-subspace division. In this study, we combine the concepts of null-subspace PRUNO and hybrid-domain ESPIRiT to provide a more robust reconstruction method that extracts null-subspace bases of calibration matrix to calculate image-domain SNMs. Multi-channel images are reconstructed by solving an image-domain nulling system formed by SNMs that contain both coil sensitivity and finite image support information, therefore, circumventing the masking-related procedure. The proposed method was evaluated with multi-channel 2D brain and knee data and compared to ESPIRiT. The proposed hybrid-domain method produced quality reconstruction highly comparable to ESPIRiT with optimal manual masking. It involved no masking-related manual procedure and was tolerant of the actual division of null- and signal-subspace. Spatial regularization could be also readily incorporated to reduce noise amplification as in ESPIRiT. We provide an efficient hybrid-domain reconstruction method using multi-channel SNMs that are calculated from coil calibration data. It eliminates the need for coil sensitivity masking and is relatively insensitive to subspace separation, therefore, presenting a robust parallel imaging reconstruction procedure in practice.

BigBrain-MR: a new digital phantom with anatomically-realistic magnetic resonance properties at 100-µm resolution for magnetic resonance methods development

The benefits, opportunities and growing availability of ultra-high field magnetic resonance imaging (MRI) for humans have prompted an expansion in research and development efforts towards increasingly more advanced high-resolution imaging techniques. To maximize their effectiveness, these efforts need to be supported by powerful computational simulation platforms that can adequately reproduce the biophysical characteristics of MRI, with high spatial resolution. In this work, we have sought to address this need by developing a novel digital phantom with realistic anatomical detail up to 100-µm resolution, including multiple MRI properties that affect image generation. This phantom, termed BigBrain-MR, was generated from the publicly available BigBrain histological dataset and lower-resolution in-vivo 7T-MRI data, using a newly-developed image processing framework that allows mapping the general properties of the latter into the fine anatomical scale of the former. Overall, the mapping framework was found to be effective and robust, yielding a diverse range of realistic “in-vivo-like” MRI contrasts and maps at 100-µm resolution. BigBrain-MR was then tested in three imaging applications (motion effects and interpolation, super-resolution imaging, and parallel imaging reconstruction) to investigate its properties, value and validity as a simulation platform. The results consistently showed that BigBrain-MR can closely approximate the behavior of real in-vivo data, more realistically and with more extensive features than a more classic option such as the Shepp-Logan phantom. Its flexibility in simulating different contrast mechanisms and artifacts may also prove valuable for educational applications. BigBrain-MR is therefore deemed a favorable choice to support methodological development and demonstration in brain MRI, and has been made freely available to the community.

Calibrationless reconstruction of uniformly-undersampled multi-channel MR data with deep learning estimated ESPIRiT maps.

To develop a truly calibrationless reconstruction method that derives An Eigenvalue Approach to Autocalibrating Parallel MRI (ESPIRiT) maps from uniformly-undersampled multi-channel MR data by deep learning. ESPIRiT, one commonly used parallel imaging reconstruction technique, forms the images from undersampled MR k-space data using ESPIRiT maps that effectively represents coil sensitivity information. Accurate ESPIRiT map estimation requires quality coil sensitivity calibration or autocalibration data. We present a U-Net based deep learning model to estimate the multi-channel ESPIRiT maps directly from uniformly-undersampled multi-channel multi-slice MR data. The model is trained using fully-sampled multi-slice axial brain datasets from the same MR receiving coil system. To utilize subject-coil geometric parameters available for each dataset, the training imposes a hybrid loss on ESPIRiT maps at the original locations as well as their corresponding locations within the standard reference multi-slice axial stack. The performance of the approach was evaluated using publicly available T1-weighed brain and cardiac data. The proposed model robustly predicted multi-channel ESPIRiT maps from uniformly-undersampled k-space data. They were highly comparable to the reference ESPIRiT maps directly computed from 24 consecutive central k-space lines. Further, they led to excellent ESPIRiT reconstruction performance even at high acceleration, exhibiting a similar level of errors and artifacts to that by using reference ESPIRiT maps. A new deep learning approach is developed to estimate ESPIRiT maps directly from uniformly-undersampled MR data. It presents a general strategy for calibrationless parallel imaging reconstruction through learning from the coil and protocol-specific data.

Accelerated SPIRiT Parallel MR Image Reconstruction Based on Joint Sparsity and Sparsifying Transform Learning

Iterative self-consistent parallel imaging reconstruction (SPIRiT) was an autocalibrating model for parallel magnetic resonance imaging reconstruction, which is often formulated as a SPIRiT reconstruction problem with some regularization terms. Some methods based on the operator splitting and alternating direction method of multipliers (ADMM) have been employed to solve the formulated regularized SPIRiT problem. In this paper, we propose to combine the sparsifying transform learning and joint sparsity with Cartesian SPIRiT parallel magnetic resonance imaging, and solve the resulting reconstruction problem by using the variable splitting and ADMM techniques. Simulation experiments on four in vivo data sets demonstrate that the proposed algorithm achieves a better image reconstruction quality than that of other competing methods. In addition, the proposed algorithm is very suitable for graphics processing unit (GPU) parallel computing, and its accelerated version, achieved by using a GPU, is very fast, requiring only 6.7 s to reconstruct a 200×200 pixel image with 8 channels.

A convex optimisation approach to parallel magnetic resonance imaging reconstruction

A convex optimisation approach to parallel magnetic resonance imaging reconstruction

Numerical and Clinical Evaluation of the Robustness of Open-source Networks for Parallel MR Imaging Reconstruction.

Deep neural networks (DNNs) for MRI reconstruction often require large datasets for training. Still, in clinical settings, the domains of datasets are diverse, and how robust DNNs are to domain differences between training and testing datasets has been an open question. Here, we numerically and clinically evaluate the generalization of the reconstruction networks across various domains under clinically practical conditions and provide practical guidance on what points to consider when selecting models for clinical application. We compare the reconstruction performance between four network models: U-Net, the deep cascade of convolutional neural networks (DC-CNNs), Hybrid Cascade, and variational network (VarNet). We used the public multicoil dataset fastMRI for training and testing and performed a single-domain test, where the domains of the dataset used for training and testing were the same, and cross-domain tests, where the source and target domains were different. We conducted a single-domain test (Experiment 1) and cross-domain tests (Experiments 2-4), focusing on six factors (the number of images, sampling pattern, acceleration factor, noise level, contrast, and anatomical structure) both numerically and clinically. U-Net had lower performance than the three model-based networks and was less robust to domain shifts between training and testing datasets. VarNet had the highest performance and robustness among the three model-based networks, followed by Hybrid Cascade and DC-CNN. Especially, VarNet showed high performance even with a limited number of training images (200 images/10 cases). U-Net was more robust to domain shifts concerning noise level than the other model-based networks. Hybrid Cascade showed slightly better performance and robustness than DC-CNN, except for robustness to noise-level domain shifts. The results of the clinical evaluations generally agreed with the results of the quantitative metrics. In this study, we numerically and clinically evaluated the robustness of the publicly available networks using the multicoil data. Therefore, this study provided practical guidance for clinical applications.