Impact of undersampling on preclinical lung T2* mapping with 3D radial UTE MRI at 7 T.

Lung diseases are almost invariably heterogeneous and progressive, making it imperative to capture temporally and spatially explicit information to understand the disease initiation and progression. Imaging the lung with MRI-particularly in the preclinical setting-has historically been challenging because of relatively low lung tissue density, rapid cardiac and respiratory motion, and rapid transverse (T2*) relaxation. These limitations can largely be mitigated using ultrashort-echo-time (UTE) sequences, which are intrinsically robust to motion and avoid significant T2* decay. A significant disadvantage of common radial UTE sequences is that they require inefficient, center-out k-space sampling, resulting in long acquisition times relative to conventional Cartesian sequences. Therefore, pulmonary images acquired with radial UTE are often undersampled to reduce acquisition time. However, undersampling reduces image SNR, introduces image artifacts, and degrades true image resolution. The level of undersampling is further increased if offline gating techniques like retrospective gating are employed, because only a portion (∼40-50%) of the data is used in the final image reconstruction. Here, we explore the impact of undersampling on SNR and T2* mapping in mouse lung imaging using simulation and in-vivo data. Increased scatter in both metrics was noticeable at around 50% sampling. Parenchymal apparent SNR only decreased slightly (average decrease âˆ¼ 1.4) with as little as 10% sampling. Apparent T2* remained similar across undersampling levels, but it became significantly increased (p < 0.05) below 80% sampling. These trends suggest that undersampling can generate quantifiable, but moderate changes in the apparent value of T2*. Moreover, these approaches to assess the impact of undersampling are straightforward to implement and can readily be expanded to assess the quantitative impact of other MR acquisition and reconstruction parameters.

Improvement of image quality for bright-blood image in VISIBLE (volume isotropic simultaneous interleaved bright- and black-blood examination) by using k-space reordering and startup echoes.

PURPOSE: A volume isotropic simultaneous interleaved bright- and black-blood examination (VISIBLE) can simultaneously acquire images with suppressed vascular signals (black-blood images) and images without suppression (bright-blood images). We aimed to improve of the bright-blood images by adjusting the k-space filling and using startup echo. METHODS: The k-space arrangement of bright-blood images in the conventional VISIBLE followed a low-to-high frequency order, whereas that in the proposed VISIBLE sequence was in the reversed order, and a startup echo was added. The effects of startup echo on the signal-to-noise ratio (SNR) were evaluated using phantoms, considering both white matter (WM) and post-contrast blood. Data from copper sulfate phantoms were acquired in 1D Fourier transform mode using both the conventional and proposed methods of the two VISIBLE sequences. The signal behavior with each sequence was evaluated. Fourteen patients with a total of 21 metastases were included in the study. For each patient, VISIBLE images of both conventional and proposed methods were obtained consecutively after the contrast agent administration. Using clinical images, we conducted a comparison of the SNR and contrast-to-noise ratio (CNR) for tumors, normal WM, and blood vessels between the conventional and proposed VISIBLE sequences. RESULTS: There was no significant difference in SNRs for both black- and bright-blood images between the conventional sequence and the proposed sequence with different number of startup echoes, however, the SNR of the proposed sequence decreased with increasing number of startup echoes in both black- and bright-images. The signal behavior of the bright-blood image reached a "steady state" when the startup echo exceeded 20. The SNRs of blood vessels in the bright-blood images did not differ significantly between conventional and proposed VISIBLE sequences. The SNRs of WM in the bright-blood images was significantly larger in the conventional sequence than in the proposed sequence. The SNRs of tumors in bright blood images was significantly larger in the proposed sequence than in the conventional sequence. The CNRs between tumors and WM, vessels and WM in the bright-blood images were significantly higher in the proposed sequence than in the conventional sequence. CONCLUSION: The use of the startup echo in combination with the high-to-low frequency k-space ordering method resulted in improved CNR of the bright-blood images in the VISIBLE sequence.

Reduction of ringing artifacts induced by diaphragm drifting in free-breathing dynamic pulmonary MRI using 3D koosh-ball acquisition.

PURPOSE: To reduce the ringing artifacts of the motion-resolved images in free-breathing dynamic pulmonary MRI. METHODS: A golden-step based interleaving (GSI) technique was proposed to reduce ringing artifacts induced by diaphragm drifting. The pulmonary MRI data were acquired using a superior-inferior navigated 3D radial UTE sequence in an interleaved manner during free breathing. Successive interleaves were acquired in an incoherent fashion along the polar direction. Four-dimensional images were reconstructed from the motion-resolved k-space data obtained by retrospectively binning. The reconstruction algorithms included standard nonuniform fast Fourier transform (NUFFT), Voronoi-density-compensated NUFFT, extra-dimensional UTE, and motion-state weighted motion-compensation reconstruction. The proposed interleaving technique was compared with a conventional sequential interleaving (SeqI) technique on a phantom and eight subjects. RESULTS: The quantified ringing artifacts level in the motion-resolved image is positively correlated with the quantified nonuniformity level of the corresponding k-space. The nonuniformity levels of the end-expiratory and end-inspiratory k-space binned from GSI data (0.34 ± 0.07, 0.33 ± 0.05) are significantly lower with statistical significance (p < 0.05) than that binned from SeqI data (0.44 ± 0.11, 0.42 ± 0.12). Ringing artifacts are substantially reduced in the dynamic images of eight subjects acquired using the proposed technique in comparison with that acquired using the conventional SeqI technique. CONCLUSION: Ringing artifacts in the motion-resolved images induced by diaphragm drifting can be reduced using the proposed GSI technique for free-breathing dynamic pulmonary MRI. This technique has the potential to reduce ringing artifacts in free-breathing liver and kidney MRI based on full-echo interleaved 3D radial acquisition.

Optimizing pseudo-spiral sampling for abdominal DCE MRI using a digital anthropomorphic phantom.

PURPOSE: For reliable DCE MRI parameter estimation, k-space undersampling is essential to meet resolution, coverage, and signal-to-noise requirements. Pseudo-spiral (PS) sampling achieves this by sampling k-space on a Cartesian grid following a spiral trajectory. The goal was to optimize PS k-space sampling patterns for abdomin al DCE MRI. METHODS: The optimal PS k-space sampling pattern was determined using an anthropomorphic digital phantom. Contrast agent inflow was simulated in the liver, spleen, pancreas, and pancreatic ductal adenocarcinoma (PDAC). A total of 704 variable sampling and reconstruction approaches were created using three algorithms using different parametrizations to control sampling density, halfscan and compressed sensing regularization. The sampling patterns were evaluated based on image quality scores and the accuracy and precision of the DCE pharmacokinetic parameters. The best and worst strategies were assessed in vivo in five healthy volunteers without contrast agent administration. The best strategy was tested in a DCE scan of a PDAC patient. RESULTS: The best PS reconstruction was found to be PS-diffuse based, with quadratic distribution of readouts on a spiral, without random shuffling, halfscan factor of 0.8, and total variation regularization of 0.05 in the spatial and temporal domains. The best scoring strategy showed sharper images with less prominent artifacts in healthy volunteers compared to the worst strategy. Our suggested DCE sampling strategy also showed high quality DCE images in the PDAC patient. CONCLUSION: Using an anthropomorphic digital phantom, we identified an optimal PS sampling strategy for abdominal DCE MRI, and demonstrated feasibility in a PDAC patient.

K-Space Approach in Optical Coherence Tomography: Rigorous Digital Transformation of Arbitrary-Shape Beams, Aberration Elimination and Super-Refocusing beyond Conventional Phase Correction Procedures.

For the most popular method of scan formation in Optical Coherence Tomography (OCT) based on plane-parallel scanning of the illuminating beam, we present a compact but rigorous K-space description in which the spectral representation is used to describe both the axial and lateral structure of the illuminating/received OCT signals. Along with the majority of descriptions of OCT-image formation, the discussed approach relies on the basic principle of OCT operation, in which ballistic backscattering of the illuminating light is assumed. This single-scattering assumption is the main limitation, whereas in other aspects, the presented approach is rather general. In particular, it is applicable to arbitrary beam shapes without the need for paraxial approximation or the assumption of Gaussian beams. The main result of this study is the use of the proposed K-space description to analytically derive a filtering function that allows one to digitally transform the initial 3D set of complex-valued OCT data into a desired (target) dataset of a rather general form. An essential feature of the proposed filtering procedures is the utilization of both phase and amplitude transformations, unlike conventionally discussed phase-only transformations. To illustrate the efficiency and generality of the proposed filtering function, the latter is applied to the mutual transformation of non-Gaussian beams and to the digital elimination of arbitrary aberrations at the illuminating/receiving aperture. As another example, in addition to the conventionally discussed digital refocusing enabling depth-independent lateral resolution the same as in the physical focus, we use the derived filtering function to perform digital "super-refocusing." The latter does not yet overcome the diffraction limit but readily enables lateral resolution several times better than in the initial physical focus.

Single-shot diffusion trace spectroscopic imaging using radial echo planar trajectories.

PURPOSE: Demonstrate the feasibility and evaluate the performance of single-shot diffusion trace-weighted radial echo planar spectroscopic imaging (Trace DW-REPSI) for quantifying the trace ADC in phantom and in vivo using a 3T clinical scanner. THEORY AND METHODS: Trace DW-REPSI datasets were acquired in 10 phantom and 10 healthy volunteers, with a maximum b-value of 1601 s/mm2 and diffusion time of 10.75 ms. The self-navigation properties of radial acquisitions were used for corrections of shot-to-shot phase and frequency shift fluctuations of the raw data. In vivo trace ADCs of total NAA (tNAA), total creatine (tCr), and total choline (tCho) extrapolated to pure gray and white matter fractions were compared, as well as trace ADCs estimated in voxels within white or gray matter-dominant regions. RESULTS: Trace ADCs in phantom show excellent agreement with reported values, and in vivo ADCs agree well with the expected differences between gray and white matter. For tNAA, tCr, and tCho, the trace ADCs extrapolated to pure gray and white matter ranged from 0.18-0.27 and 0.26-0.38 µm2/ms, respectively. In sets of gray and white matter-dominant voxels, the values ranged from 0.21 to 0.27 and 0.24 to 0.31 µm2/ms, respectively. The overestimated trace ADCs from this sequence can be attributed to the short diffusion time. CONCLUSION: This study presents the first demonstration of the single-shot diffusion trace-weighted spectroscopic imaging sequence using radial echo planar trajectories. The Trace DW-REPSI sequence could provide an estimate of the trace ADC in a much shorter scan time compared to conventional approaches that require three separate measurements.

Unsupervised MRI motion artifact disentanglement: introducing MAUDGAN.

Objective.This study developed an unsupervised motion artifact reduction method for magnetic resonance imaging (MRI) images of patients with brain tumors. The proposed novel design uses multi-parametric multicenter contrast-enhanced T1W (ceT1W) and T2-FLAIR MRI images.Approach.The proposed framework included two generators, two discriminators, and two feature extractor networks. A 3-fold cross-validation was used to train and fine-tune the hyperparameters of the proposed model using 230 brain MRI images with tumors, which were then tested on 148 patients'in-vivodatasets. An ablation was performed to evaluate the model's compartments. Our model was compared with Pix2pix and CycleGAN. Six evaluation metrics were reported, including normalized mean squared error (NMSE), structural similarity index (SSIM), multi-scale-SSIM (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), and multi-scale gradient magnitude similarity deviation (MS-GMSD). Artifact reduction and consistency of tumor regions, image contrast, and sharpness were evaluated by three evaluators using Likert scales and compared with ANOVA and Tukey's HSD tests.Main results.On average, our method outperforms comparative models to remove heavy motion artifacts with the lowest NMSE (18.34±5.07%) and MS-GMSD (0.07 ± 0.03) for heavy motion artifact level. Additionally, our method creates motion-free images with the highest SSIM (0.93 ± 0.04), PSNR (30.63 ± 4.96), and VIF (0.45 ± 0.05) values, along with comparable MS-SSIM (0.96 ± 0.31). Similarly, our method outperformed comparative models in removingin-vivomotion artifacts for different distortion levels except for MS- SSIM and VIF, which have comparable performance with CycleGAN. Moreover, our method had a consistent performance for different artifact levels. For the heavy level of motion artifacts, our method got the highest Likert scores of 2.82 ± 0.52, 1.88 ± 0.71, and 1.02 ± 0.14 (p-values≪0.0001) for our method, CycleGAN, and Pix2pix respectively. Similar trends were also found for other motion artifact levels.Significance.Our proposed unsupervised method was demonstrated to reduce motion artifacts from the ceT1W brain images under a multi-parametric framework.

Simulation of time-intensity curve based on k-space filling in breast dynamic contrast-enhanced three-dimensional magnetic resonance imaging.

This study elucidated the effects of a three-dimensional k-space trajectory incorporating the partial Fourier (PF) technique on a time-intensity curve (TIC) in a dynamic contrast-enhanced magnetic resonance imaging of a typical malignant breast tumor using a digital phantom. Images were obtained from the Cancer Imaging Archive Open Data for Breast Cancer, and 1-min scans with high temporal resolution were analyzed. The order of the k-space trajectory was set as Linear (sequential), Low-High (centric), PF (62.5%; Z-, Y-, and both directions), and Low-High Radial. k0 (center of the k-space) timing and TIC shape were affected by the chosen k-space trajectory and implementation of the PF technique. A small TIC gradient was obtained using a Low-High Radial order. If the k-space filling method (particularly the radial method) produces a gentle TIC gradient, misinterpretation could arise during the assessment of tumor malignancy status.

3D whole heart k-space-based super-resolution cardiac T1 mapping using rotated stacks.

Objective. To provide three-dimensional (3D) whole-heart high-resolution isotropic cardiac T1 maps using a k-space-based through-plane super-resolution reconstruction (SRR) with rotated multi-slice stacks.Approach. Due to limited SNR and cardiac motion, often only 2D T1 maps with low through-plane resolution (4-8 mm) can be obtained. Previous approaches used SRR to calculate 3D high-resolution isotropic cardiac T1 maps. However, they were limited to the ventricles. The proposed approach acquires rotated stacks in long-axis orientation with high in-plane resolution but low through-plane resolution. This results in radially overlapping stacks from which high-resolution T1 maps of the whole heart are reconstructed using a k-space-based SRR framework considering the complete acquisition model. Cardiac and residual respiratory motion between different breath holds is estimated and incorporated into the reconstruction. The proposed approach was evaluated in simulations and phantom experiments and successfully applied to ten healthy subjects.Main results. 3D T1 maps of the whole heart were obtained in the same acquisition time as previous methods covering only the ventricles. T1 measurements were possible even for small structures, such as the atrial wall. The proposed approach provided accurate (P> 0.4;R2> 0.99) and precise T1 values (SD of 64.32 ± 22.77 ms in the proposed approach, 44.73 ± 31.9 ms in the reference). The edge sharpness of the T1 maps was increased by 6.20% and 4.73% in simulation and phantom experiments, respectively. Contrast-to-noise ratios between the septum and blood pool increased by 14.50% inin vivomeasurements with a k-space compared to an image-space-based SRR.Significance. The proposed approach provided whole-heart high-resolution 1.3 mm isotropic T1 maps in an overall acquisition time of approximately three minutes. Small structures, such as the atrial and right ventricular walls, could be visualized in the T1 maps.

Impact of Deep Learning Image Reconstruction Methods on MRI Throughput.

Purpose To evaluate the effect of implementing two distinct commercially available deep learning reconstruction (DLR) algorithms on the efficiency of MRI examinations conducted in real clinical practice within an outpatient setting at a large, multicenter institution. Materials and Methods This retrospective study included 7346 examinations from 10 clinical MRI scanners analyzed during the pre- and postimplementation periods of DLR methods. Two different types of DLR methods, namely Digital Imaging and Communications in Medicine (DICOM)-based and k-space-based methods, were implemented in half of the scanners (three DICOM-based and two k-space-based), while the remaining five scanners had no DLR method implemented. Scan and room times of each examination type during the pre- and postimplementation periods were compared among the different DLR methods using the Wilcoxon test. Results The application of deep learning methods resulted in significant reductions in scan and room times for certain examination types. The DICOM-based method demonstrated up to a 53% reduction in scan times and a 41% reduction in room times for various study types. The k-space-based method demonstrated up to a 27% reduction in scan times but did not significantly reduce room times. Conclusion DLR methods were associated with reductions in scan and room times in a clinical setting, though the effects were heterogeneous depending on examination type. Thus, potential adopters should carefully evaluate their case mix to determine the impact of integrating these tools. Keywords: Deep Learning MRI Reconstruction, Reconstruction Algorithms, DICOM-based Reconstruction, k-Space-based Reconstruction © RSNA, 2024 See also the commentary by GharehMohammadi in this issue.

Ultra-rapid, Free-breathing, Real-time Cardiac Cine MRI Using GRASP Amplified with View Sharing and KWIC Filtering.

Purpose To achieve ultra-high temporal resolution (approximately 20 msec) in free-breathing, real-time cardiac cine MRI using golden-angle radial sparse parallel (GRASP) reconstruction amplified with view sharing (VS) and k-space-weighted image contrast (KWIC) filtering. Materials and Methods Fourteen pediatric patients with congenital heart disease (mean age [SD], 9 years ± 2; 13 male) and 10 adult patients with arrhythmia (mean age, 62 years ± 8; nine male) who underwent both standard breath-hold cine and free-breathing real-time cine using GRASP were retrospectively identified. To achieve high temporal resolution, each time frame was reconstructed using six radial spokes, corresponding to acceleration factors ranging from 24 to 32. To compensate for loss in spatial resolution resulting from over-regularization in GRASP, VS and KWIC filtering were incorporated. The blur metric, visual image quality scores, and biventricular parameters were compared between clinical and real-time cine images. Results In pediatric patients, the incorporation of VS and KWIC into GRASP (ie, GRASP + VS + KWIC) produced significantly (P < .05) sharper x-y-t (blur metric: 0.36 ± 0.03, 0.41 ± 0.03, 0.48 ± 0.03, respectively) and x-y-f (blur metric: 0.28 ± 0.02, 0.31 ± 0.03, 0.37 ± 0.03, respectively) component images compared with GRASP + VS and conventional GRASP. Only the noise score differed significantly between GRASP + VS + KWIC and clinical cine; all visual scores were above the clinically acceptable (3.0) cutoff point. Biventricular volumetric parameters strongly correlated (R2 > 0.85) between clinical and real-time cine images reconstructed with GRASP + VS + KWIC and were in good agreement (relative error < 6% for all parameters). In adult patients, the visual scores of all categories were significantly lower (P < .05) for clinical cine compared with real-time cine with GRASP + VS + KWIC, except for noise (P = .08). Conclusion Incorporating VS and KWIC filtering into GRASP reconstruction enables ultra-high temporal resolution (approximately 20 msec) without significant loss in spatial resolution. Keywords: Cine, View Sharing, k-Space-weighted Image Contrast Filtering, Radial k-Space, Pediatrics, Arrhythmia, GRASP, Compressed Sensing, Real-Time, Free-Breathing Supplemental material is available for this article. © RSNA, 2024.

Variable augmentation network for invertible MR coil compression.

To improve the efficiency of multi-coil data compression and recover the compressed image reversibly, increasing the possibility of applying the proposed method to medical scenarios. A deep learning algorithm is employed for MR coil compression in the presented work. The approach introduces a variable augmentation network for invertible coil compression (VAN-ICC). This network utilizes the inherent reversibility of normalizing flow-based models. The aim is to enhance the readability of the sentence and clearly convey the key components of the algorithm. By applying the variable augmentation technology to image/k-space variables from multi-coils, VAN-ICC trains the invertible network by finding an invertible and bijective function, which can map the original data to the compressed counterpart and vice versa. Experiments conducted on both fully-sampled and under-sampled data verified the effectiveness and flexibility of VAN-ICC. Quantitative and qualitative comparisons with traditional non-deep learning-based approaches demonstrated that VAN-ICC carries much higher compression effects. The proposed method trains the invertible network by finding an invertible and bijective function, which improves the defects of traditional coil compression method by utilizing inherent reversibility of normalizing flow-based models. In addition, the application of variable augmentation technology ensures the implementation of reversible networks. In short, VAN-ICC offered a competitive advantage over other traditional coil compression algorithms.

A novel automatic segmentation method directly based on magnetic resonance imaging K-space data for auxiliary diagnosis of glioma.

Background: The use of segmentation architectures in medical imaging, particularly for glioma diagnosis, marks a significant advancement in the field. Traditional methods often rely on post-processed images; however, key details can be lost during the fast Fourier transformation (FFT) process. Given the limitations of these techniques, there is a growing interest in exploring more direct approaches. The adaption of segmentation architectures originally designed for road extraction for medical imaging represents an innovative step in this direction. By employing K-space data as the modal input, this method completely eliminates the information loss inherent in FFT, thereby potentially enhancing the precision and effectiveness of glioma diagnosis. Methods: In the study, a novel architecture based on a deep-residual U-net was developed to accomplish the challenging task of automatically segmenting brain tumors from K-space data. Brain tumors from K-space data with different under-sampling rates were also segmented to verify the clinical application of our method. Results: Compared to the benchmarks set in the 2018 Brain Tumor Segmentation (BraTS) Challenge, our proposed architecture had superior performance, achieving Dice scores of 0.8573, 0.8789, and 0.7765 for the whole tumor (WT), tumor core (TC), and enhanced tumor (ET) regions, respectively. The corresponding Hausdorff distances were 2.5649, 1.6146, and 2.7187 for the WT, TC, and ET regions, respectively. Notably, compared to traditional image-based approaches, the architecture also exhibited an improvement of approximately 10% in segmentation accuracy on the K-space data at different under-sampling rates. Conclusions: These results show the superiority of our method compared to previous methods. The direct performance of lesion segmentation based on K-space data eliminates the time-consuming and tedious image reconstruction process, thus enabling the segmentation task to be accomplished more efficiently.

On retrospective k-space subsampling schemes for deep MRI reconstruction.

Acquiring fully-sampled MRI k-space data is time-consuming, and collecting accelerated data can reduce the acquisition time. Employing 2D Cartesian-rectilinear subsampling schemes is a conventional approach for accelerated acquisitions; however, this often results in imprecise reconstructions, even with the use of Deep Learning (DL), especially at high acceleration factors. Non-rectilinear or non-Cartesian trajectories can be implemented in MRI scanners as alternative subsampling options. This work investigates the impact of the k-space subsampling scheme on the quality of reconstructed accelerated MRI measurements produced by trained DL models. The Recurrent Variational Network (RecurrentVarNet) was used as the DL-based MRI-reconstruction architecture. Cartesian, fully-sampled multi-coil k-space measurements from three datasets were retrospectively subsampled with different accelerations using eight distinct subsampling schemes: four Cartesian-rectilinear, two Cartesian non-rectilinear, and two non-Cartesian. Experiments were conducted in two frameworks: scheme-specific, where a distinct model was trained and evaluated for each dataset-subsampling scheme pair, and multi-scheme, where for each dataset a single model was trained on data randomly subsampled by any of the eight schemes and evaluated on data subsampled by all schemes. In both frameworks, RecurrentVarNets trained and evaluated on non-rectilinearly subsampled data demonstrated superior performance, particularly for high accelerations. In the multi-scheme setting, reconstruction performance on rectilinearly subsampled data improved when compared to the scheme-specific experiments. Our findings demonstrate the potential for using DL-based methods, trained on non-rectilinearly subsampled measurements, to optimize scan time and image quality.

Incorporation of view sharing and KWIC filtering into GRASP-Pro improves spatial resolution of single-shot, multi-TI, late gadolinium enhancement MRI.

While single-shot late gadolinium enhancement (LGE) is useful for imaging patients with arrhythmia and/or dyspnea, it produces low spatial resolution. One approach to improve spatial resolution is to accelerate data acquisition using compressed sensing (CS). Our previous work described a single-shot, multi-inversion time (TI) LGE pulse sequence using radial k-space sampling and CS, but over-regularization resulted in significant image blurring that muted the benefits of data acceleration. The purpose of the present study was to improve the spatial resolution of the single-shot, multi-TI LGE pulse sequence by incorporating view sharing (VS) and k-space weighted contrast (KWIC) filtering into a GRASP-Pro reconstruction. In 24 patients (mean age = 61 ± 16 years; 9/15 females/males), we compared the performance of our improved multi-TI LGE and standard multi-TI LGE, where clinical standard LGE was used as a reference. Two clinical raters independently graded multi-TI images and clinical LGE images visually on a five-point Likert scale (1, nondiagnostic; 3, clinically acceptable; 5, best) for three categories: the conspicuity of myocardium or scar, artifact, and noise. The summed visual score (SVS) was defined as the sum of the three scores. Myocardial scar volume was quantified using the full-width at half-maximum method. The SVS was not significantly different between clinical breath-holding LGE (median 13.5, IQR 1.3) and multi-TI LGE (median 12.5, IQR 1.6) (P = 0.068). The myocardial scar volumes measured from clinical standard LGE and multi-TI LGE were strongly correlated (coefficient of determination, R2 = 0.99) and in good agreement (mean difference = 0.11%, lower limit of the agreement = -2.13%, upper limit of the agreement = 2.34%). The inter-rater agreement in myocardial scar volume quantification was strong (intraclass correlation coefficient = 0.79). The incorporation of VS and KWIC into GRASP-Pro improved spatial resolution. Our improved 25-fold accelerated, single-shot LGE sequence produces clinically acceptable image quality, multi-TI reconstruction, and accurate myocardial scar volume quantification.

MA-RECON: Mask-aware deep-neural-network for robust fast MRI k-space interpolation.

BACKGROUND AND OBJECTIVE: High-quality reconstruction of MRI images from under-sampled 'k-space' data, which is in the Fourier domain, is crucial for shortening MRI acquisition times and ensuring superior temporal resolution. Over recent years, a wealth of deep neural network (DNN) methods have emerged, aiming to tackle the complex, ill-posed inverse problem linked to this process. However, their instability against variations in the acquisition process and anatomical distribution exposes a deficiency in the generalization of relevant physical models within these DNN architectures. The goal of our work is to enhance the generalization capabilities of DNN methods for k-space interpolation by introducing 'MA-RECON', an innovative mask-aware DNN architecture and associated training method. METHODS: Unlike preceding approaches, our 'MA-RECON' architecture encodes not only the observed data but also the under-sampling mask within the model structure. It implements a tailored training approach that leverages data generated with a variety of under-sampling masks to stimulate the model's generalization of the under-sampled MRI reconstruction problem. Therefore, effectively represents the associated inverse problem, akin to the classical compressed sensing approach. RESULTS: The benefits of our MA-RECON approach were affirmed through rigorous testing with the widely accessible fastMRI dataset. Compared to standard DNN methods and DNNs trained with under-sampling mask augmentation, our approach demonstrated superior generalization capabilities. This resulted in a considerable improvement in robustness against variations in both the acquisition process and anatomical distribution, especially in regions with pathology. CONCLUSION: In conclusion, our mask-aware strategy holds promise for enhancing the generalization capacity and robustness of DNN-based methodologies for MRI reconstruction from undersampled k-space data. Code is available in the following link: https://github.com/nitzanavidan/PD_Recon.

McSTRA: A multi-branch cascaded swin transformer for point spread function-guided robust MRI reconstruction.

Deep learning MRI reconstruction methods are often based on Convolutional neural network (CNN) models; however, they are limited in capturing global correlations among image features due to the intrinsic locality of the convolution operation. Conversely, the recent vision transformer models (ViT) are capable of capturing global correlations by applying self-attention operations on image patches. Nevertheless, the existing transformer models for MRI reconstruction rarely leverage the physics of MRI. In this paper, we propose a novel physics-based transformer model titled, the Multi-branch Cascaded Swin Transformers (McSTRA) for robust MRI reconstruction. McSTRA combines several interconnected MRI physics-related concepts with the Swin transformers: it exploits global MRI features via the shifted window self-attention mechanism; it extracts MRI features belonging to different spectral components via a multi-branch setup; it iterates between intermediate de-aliasing and data consistency via a cascaded network with intermediate loss computations; furthermore, we propose a point spread function-guided positional embedding generation mechanism for the Swin transformers which exploit the spread of the aliasing artifacts for effective reconstruction. With the combination of all these components, McSTRA outperforms the state-of-the-art methods while demonstrating robustness in adversarial conditions such as higher accelerations, noisy data, different undersampling protocols, out-of-distribution data, and abnormalities in anatomy.

MRI motion artifact reduction using a conditional diffusion probabilistic model (MAR-CDPM).

BACKGROUND: High-resolution magnetic resonance imaging (MRI) with excellent soft-tissue contrast is a valuable tool utilized for diagnosis and prognosis. However, MRI sequences with long acquisition time are susceptible to motion artifacts, which can adversely affect the accuracy of post-processing algorithms. PURPOSE: This study proposes a novel retrospective motion correction method named "motion artifact reduction using conditional diffusion probabilistic model" (MAR-CDPM). The MAR-CDPM aimed to remove motion artifacts from multicenter three-dimensional contrast-enhanced T1 magnetization-prepared rapid acquisition gradient echo (3D ceT1 MPRAGE) brain dataset with different brain tumor types. MATERIALS AND METHODS: This study employed two publicly accessible MRI datasets: one containing 3D ceT1 MPRAGE and 2D T2-fluid attenuated inversion recovery (FLAIR) images from 230 patients with diverse brain tumors, and the other comprising 3D T1-weighted (T1W) MRI images of 148 healthy volunteers, which included real motion artifacts. The former was used to train and evaluate the model using the in silico data, and the latter was used to evaluate the model performance to remove real motion artifacts. A motion simulation was performed in k-space domain to generate an in silico dataset with minor, moderate, and heavy distortion levels. The diffusion process of the MAR-CDPM was then implemented in k-space to convert structure data into Gaussian noise by gradually increasing motion artifact levels. A conditional network with a Unet backbone was trained to reverse the diffusion process to convert the distorted images to structured data. The MAR-CDPM was trained in two scenarios: one conditioning on the time step t $t$ of the diffusion process, and the other conditioning on both t $t$ and T2-FLAIR images. The MAR-CDPM was quantitatively and qualitatively compared with supervised Unet, Unet conditioned on T2-FLAIR, CycleGAN, Pix2pix, and Pix2pix conditioned on T2-FLAIR models. To quantify the spatial distortions and the level of remaining motion artifacts after applying the models, quantitative metrics were reported including normalized mean squared error (NMSE), structural similarity index (SSIM), multiscale structural similarity index (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), and multiscale gradient magnitude similarity deviation (MS-GMSD). Tukey's Honestly Significant Difference multiple comparison test was employed to quantify the difference between the models where p-value  < 0.05 $ < 0.05$ was considered statistically significant. RESULTS: Qualitatively, MAR-CDPM outperformed these methods in preserving soft-tissue contrast and different brain regions. It also successfully preserved tumor boundaries for heavy motion artifacts, like the supervised method. Our MAR-CDPM recovered motion-free in silico images with the highest PSNR and VIF for all distortion levels where the differences were statistically significant (p-values < 0.05 $< 0.05$ ). In addition, our method conditioned on t and T2-FLAIR outperformed (p-values < 0.05 $< 0.05$ ) other methods to remove motion artifacts from the in silico dataset in terms of NMSE, MS-SSIM, SSIM, and MS-GMSD. Moreover, our method conditioned on only t outperformed generative models (p-values < 0.05 $< 0.05$ ) and had comparable performances compared with the supervised model (p-values > 0.05 $> 0.05$ ) to remove real motion artifacts. CONCLUSIONS: The MAR-CDPM could successfully remove motion artifacts from 3D ceT1 MPRAGE. It is particularly beneficial for elderly who may experience involuntary movements during high-resolution MRI imaging with long acquisition times.

Fast 19 F spectroscopic imaging with pseudo-spiral k-space sampling.

Fluorine MRI is finding wider acceptance in theranostics applications where imaging of 19 F hotspots of fluorinated contrast material is central. The essence of such applications is to capture ghosting-artifact-free images of the inherently low MR response under clinically viable conditions. To serve this purpose, this work introduces the balanced spiral spectroscopic imaging (BaSSI) sequence, which is implemented on a 3.0 T clinical scanner and is capable of generating 19 F hotspot images in an efficient manner. The sequence utilizes an all-phase-encoded pseudo-spiral k-space trajectory, enabling the acquisition of broadband (80 ppm) fluorine spectra free from chemical shift ghosting. BaSSI can acquire a 64 × 64 image with 1 mm × 1 mm voxels in just 14 s, significantly outperforming typical MRSI sequences used in 1 H or 31 P imaging. The study employed in silico characterization to verify essential design choices such as the excitation pulse, as well as to identify the boundaries of the parameter space explored for optimization. BaSSI's performance was further benchmarked against the 3D ultrashort-echo-time balanced steady-state free precession (3D UTE BSSFP) sequence, a well established method used in 19 F MRI, in vitro. Both sequences underwent extensive optimization through exploration of a wide parameter space on a small phantom containing 10 µL of non-diluted bulk perfluorooctylbromide (PFOB) prior to comparative experiments. Subsequent to optimization, BaSSI and 3D UTE BSSFP were employed to capture images of small non-diluted bulk PFOB samples (0.10 and 0.05 µL), with variations in the number of signal averages, and thus the total scan time, in order to assess the detection sensitivities of the sequences. In these experiments, the detection sensitivity was evaluated using the Rose criterion (Rc ), which provides a quantitative metric for assessing object visibility. The study further demonstrated BaSSI's utility as a (pre)clinical tool through postmortem imaging of polymer microspheres filled with PFOB in a BALB/c mouse. Anatomic localization of 19 F hotspots was achieved by denoising raw data obtained with BaSSI using a filter based on the Rose criterion. These data were then successfully registered to 1 H anatomical images. BaSSI demonstrated superior detection sensitivity in the benchmarking analysis, achieving Rc values approximately twice as high as those obtained with the 3D UTE BSSFP method. The technique successfully facilitated imaging and precise localization of 19 F hotspots in postmortem experiments. However, it is important to highlight that imaging 10 mM PFOB in small mice postmortem, utilizing a 48 × 48 × 48 3D scan, demanded a substantial scan time of 1 h and 45 min. Further studies will explore accelerated imaging techniques, such as compressed sensing, to enhance BaSSI's clinical utility.

k-strip: A novel segmentation algorithm in k-space for the application of skull stripping.

BACKGROUND AND OBJECTIVE: We present a novel deep learning-based skull stripping algorithm for magnetic resonance imaging (MRI) that works directly in the information rich complex valued k-space. METHODS: Using four datasets from different institutions with a total of around 200,000 MRI slices, we show that our network can perform skull-stripping on the raw data of MRIs while preserving the phase information which no other skull stripping algorithm is able to work with. For two of the datasets, skull stripping performed by HD-BET (Brain Extraction Tool) in the image domain is used as the ground truth, whereas the third and fourth dataset comes with per-hand annotated brain segmentations. RESULTS: All four datasets were very similar to the ground truth (DICE scores of 92 %-99 % and Hausdorff distances of under 5.5 pixel). Results on slices above the eye-region reach DICE scores of up to 99 %, whereas the accuracy drops in regions around the eyes and below, with partially blurred output. The output of k-Strip often has smoothed edges at the demarcation to the skull. Binary masks are created with an appropriate threshold. CONCLUSION: With this proof-of-concept study, we were able to show the feasibility of working in the k-space frequency domain, preserving phase information, with consistent results. Besides preserving valuable information for further diagnostics, this approach makes an immediate anonymization of patient data possible, already before being transformed into the image domain. Future research should be dedicated to discovering additional ways the k-space can be used for innovative image analysis and further workflows.