To develop a data-efficient strategy for accelerated MRI reconstruction with Diffusion Probabilistic Generative Models (DPMs) that enables faster scan times in clinical stroke MRI when only limited fully-sampled data are available. Our simple training strategy first pre-trains a DPM on a large, diverse collection of publicly available fastMRI brain data and then fine-tunes on a small target dataset using carefully selected learning rates and fine-tuning durations. The approach is evaluated on controlled fastMRI experiments and on clinical stroke MRI data with a blinded clinical reader study. DPMs pre-trained on 4000 non-FLAIR subjects and fine-tuned on FLAIR data from only 20 target subjects achieve reconstruction performance comparable to models trained with substantially more target-domain FLAIR data across multiple acceleration factors. Moderate fine-tuning with a reduced learning rate yields improved performance, while insufficient or excessive fine-tuning degrades reconstruction quality. In a blinded reader study of 80 subjects at a single clinical site, images reconstructed from 2 × $$ 2\times $$ accelerated data using the proposed approach are rated comparably to standard-of-care on the image quality and structural delineation metrics defined in this work. Large-scale pre-training combined with targeted fine-tuning can enable DPM-based MRI reconstruction for our data-constrained, accelerated clinical stroke MRI application. In the single-site settings evaluated here, the proposed approach reduces the need for large application-specific datasets while maintaining clinically acceptable image quality, providing preliminary evidence for pre-trained and fine-tuned diffusion models as a strategy for accelerated MRI in targeted applications.
Magnetic resonance imaging (MRI) is a non-invasive method that relies on a highly homogeneous B0 magnetic field to generate high-quality images. However, subject-specific inhomogeneities, induced by magnetic susceptibility differences, can degrade image quality and cause artifacts and are often inadequately corrected by traditional active or passive shimming methods. This work aims to create a new passive shimming method to overcome these problems. This study introduces an innovative approach to developing custom, subject-specific passive shims 3D-printed with a binder and a ferromagnetic ink. The optimal spatial distribution and concentration of ferromagnetic material within the shim are calculated using a custom algorithm that minimizes the magnetic field standard deviation. The shim design and fabrication process is demonstrated for in vivo passive shimming of a rat brain, yielding a 21% improvement in magnetic field homogeneity and improved image quality in spin-echo echo-planar imaging scans, with a 65% increase in the Dice-Sørensen similarity coefficient. This approach highlights the potential of combining computational optimization with additive manufacturing for effective, subject-specific passive shimming in MRI.
Magnetic Resonance Fingerprinting (MRF) enables rapid quantitative imaging, but high-resolution 3D reconstructions remain computationally expensive due to the NUFFTs required at every iteration, and the commonly used Locally Low Rank (LLR) regularization becomes ineffective at high acceleration. Learned 3D priors could address these limitations, but training them at scale is challenging due to memory and runtime constraints. This work proposes SPUR-iG, a fully 3D deep unrolled subspace reconstruction framework that provides fast, high-quality reconstruction for high-resolution non-Cartesian 3D MRF, while keeping training time computationally tractable. SPUR-iG leverages implicit GROG-based data consistency (DC), which grids non-Cartesian k-space using a learned family of kernels, enabling efficient FFT-based DC with minimal artifacts. To make 3D unrolled training more efficient, we introduce a staged training strategy that keeps computation tractable while progressively improving reconstruction quality. We evaluate the method on a large in vivo dataset, as well as on cross-vendor out-of-distribution data. At 1 mm isotropic resolution, SPUR-iG outperforms LLR and a state-of-the-art hybrid 2D-3D unrolled baseline in subspace coefficient quality and T 1 / T 2 accuracy. Whole-brain reconstructions complete in under 15 s, providing up to a 111 × speedup for 2-min scans relative to LLR. Notably, SPUR-iG reconstructions from 30-s acquisitions achieve mean T 1 accuracy that matches or exceeds the mean accuracy of LLR reconstructions from 2-min acquisitions. SPUR-iG introduces a fully 3D unrolled reconstruction framework for MRF that improves both reconstruction speed and accuracy, making high-resolution accelerated 3D MRF more practical for research and clinical use.
This study aims to address the suppression of oscillation and ghosting appearing in RF-spoiled gradient-echo-based dynamic imaging, which is caused by the pseudo-steady state (PSS) of magnetization. We extended the concept of the previous RF phase-cycle-adapted averaging method to RF-spoiled gradient-echo-based cine (or 2D line-scan) imaging by introducing an appropriate intertrial RF phase-cycle shift while maintaining flexibility in the number of phase-encoding lines. The proposed method was validated through theoretical analysis based on the PSS magnetization, 2D Bloch simulations, phantom and human brain imaging on a 3 T clinical scanner. Furthermore, its performance was evaluated using spectral analysis and tSNR maps. In both phantom and human brain images, the proposed method effectively suppressed ghosting artifacts and oscillatory components at specific frequencies. Averaging over multiple trials, whose number is determined by the number of phase increment (ψ) rotations required for the PSS magnetization to return to its initial phase, fully eliminated the ghosts and oscillatory components, and even when the number of trials was reduced, it significantly mitigated them. The proposed method provided robust performance in both phantom and human brain images, with improved signal quality as shown in tSNR maps. In this work, we theoretically investigated the PSS behavior of magnetization in RF-spoiled gradient-echo-based cine imaging and demonstrated a method to effectively suppress oscillation and ghosting appearing in RF-spoiled gradient-echo-based cine images. This study is expected to provide a valuable tool for various applications of ultrahigh temporal resolution MRI, including fMRI, utilizing cine imaging.
To develop SAFE, a self-calibrated framework to estimate B1 + and B0 field inhomogeneities directly from conventional magnetic resonance fingerprinting (MRF) acquisitions and improve its quantification accuracy. SAFE utilized a two-step approach. First, two physics-informed image markers are extracted from the MRF data to create a magnitude and a phase image that are highly correlated with B1 + and B0 inhomogeneities, respectively. Second, a deep learning (DL) network is applied to map these markers to quantitative field maps. SAFE was tested on 3D Spiral Projection Imaging MRF brain acquisition at 3T, where the network was trained and validated across a multi-site, multi-vendor dataset (N = 358) from healthy volunteers and clinical population. The capability of SAFE to be applied to unseen MRF sequences with different signal preparations and flip-angle trains through adaptively retraining without the need for additional training data was also tested. SAFE achieved normalized-root-mean-square-error within 3% against gold-standard field-calibration scans on both B1 + and B0 maps (N = 32), with high performance remaining on datasets from scanners that the training data were acquired from. T1 and T2 biases were shown to be effectively corrected on healthy volunteers, a patient with brain tumor, and a pediatric subject. Tissue quantification accuracy after SAFE-estimated field correction was validated on a large patient cohort (N = 86). SAFE was also demonstrated to be adaptable to different MRF sequences without acquiring additional training data. By combining physics-informed image markers with DL, the proposed SAFE framework enables calibration-free estimation of B1 + and B0 field inhomogeneities at 3T for whole brain MRF.
To develop a unified image reconstruction framework that bridges real-time and gated cardiac MRI, including quantitative MRI. We introduce generative multitasking, which learns subject- and dataset-specific implicit neural temporal bases from sequence timings and an interpretable latent space for cardiac and respiratory motion. Cardiac motion is modeled as a complex harmonic, with phase encoding timing and a latent amplitude capturing beat-to-beat functional variability, linking cardiac phase-resolved ("gated-like") and time-resolved ("real-time-like") views. We implemented the framework using a conditional variational autoencoder (CVAE) and evaluated it for free-breathing, non-ECG-gated radial GRE in three settings: steady-state cine imaging, multicontrast T2prep/inversion-recovery imaging, and dual-flip-angle T1/T2 mapping, compared with conventional multitasking. Generative multitasking provided flexible cardiac motion representation, enabling reconstruction of archetypal cardiac phase-resolved cines (like gating) as well as time-resolved series that reveal beat-to-beat variability (like real-time imaging). Conditioning on the previous k-space angle and modifying this term at inference removed eddy-current artifacts without globally smoothing high temporal frequencies. For quantitative mapping, generative multitasking reduced intraseptal T1 and T2 coefficients of variation (CoV) compared with conventional multitasking (T1: 0.13 vs. 0.31; T2: 0.12 vs. 0.32; p < 0.001), indicating higher SNR. Generative multitasking uses a CVAE with complex harmonic cardiac coordinates to unify gated and real-time CMR within a single free-breathing, non-ECG-gated acquisition. The framework allows flexible cardiac motion representation, suppresses trajectory-dependent artifacts, and improves T1 and T2 mapping, suggesting a path toward cine, multicontrast, and quantitative imaging without separate gated and real-time scans.
To enhance the accuracy and reproducibility for in vivo taurine quantification using the double-quantum filtering (DQF) technique. Simulations and phantom experiments were conducted at both 7 T and 3 T to determine the optimal parameter τ by calculating the taurine purity index among other neighboring metabolites. The frequency-selective read-pulse bandwidth and the dephasing/rephasing gradient moments were further adjusted to reduce residual water side band interference. Rat models of C6 gliomas were compared to normal brain for taurine level at 7 T. Metabolomics was conducted to validate the accuracy of taurine quantification. Human gray and white matter taurine differences were revealed at 3 T. Test-retest reliability was evaluated on humans. Simulations and phantom experiments identified optimal τ for maximizing taurine signal purity as 102 ms at 7 T and 92 ms at 3 T. The best combinations of parameters for minimal water interference were determined at 3 T. Optimized DQF sequences yielded lower CRLBs. In rat brains, taurine concentration measured with optimized DQF sequences exhibited strong correlation with metabolomics, outperforming both conventional MRS and nonoptimized DQF sequences. Only the optimized method revealed significant taurine concentration differences between gray and white matter. Test-retest demonstrated good reproducibility in gray matter (ICC = 0.82) for the optimized method. The optimized DQF method enhances taurine detection accuracy and robustness in both animal and human studies and was for the first time applied in human brain taurine test at 3 T.
QQ, a recently proposed oxygen extraction fraction (OEF) mapping technique combining quantitative susceptibility mapping (QSM) and quantitative blood oxygen level-dependent (qBOLD) (QSM + qBOLD = QQ), generates OEF maps noninvasively from a single routine MRI sequence, without requiring vascular challenges used in other OEF approaches. A deep learning approach, QQ-NET, further enables rapid 3D OEF reconstruction (˜1.5 min), but it is trained on a fixed echo-time (TE) scheme and must be retrained whenever acquisition protocols differ, limiting its clinical applicability. This study introduces QQ-F, a novel deep learning approach designed to eliminate the need for retraining. QQ-F incorporates a feature extraction unit that derives QQ model-related features as inputs, rather than relying directly on raw signals. For a fair comparison, QQ-F was trained using the same 3D multi-echo gradient echo (mGRE) dataset as QQ-NET, acquired from 26 ischemic stroke patients. Both models were tested using simulations and data from 24 multiple sclerosis (MS) and 30 dementia patients acquired with varying TE sequences. In simulations, QQ-F provided more accurate OEF maps than QQ-NET with lower mean absolute error. In patient datasets-particularly dementia datasets, where TE values differed substantially from QQ-NET's training protocol-QQ-F yielded significantly higher lesion-to-normal tissue contrast than QQ-NET, indicating superior robustness to acquisition variability. QQ-F enables deep learning-based QQ OEF mapping across diverse MR acquisition protocols without retraining, thereby enhancing the clinical scalability of QQ-based OEF mapping.
Recent updates to the diagnostic criteria of multiple sclerosis (MS) require whole-brain T2*-weighted (T2*w) imaging with submillimeter resolution to detect novel diagnostic biomarkers such as the central vein sign. However, to achieve the needed submillimeter spatial resolution, conventional T2*w 3D gradient-echo scans sequences are limited by prohibitively long scan times for clinical use. Here, we evaluated a different approach based on a segmented 3D echo planar imaging (3D-EPI) sequence, accelerated with 2D Controlled Aliasing in Parallel Imaging Results in Higher Acceleration (CAIPIRINHA) undersampling and denoised with a deep learning-based network. Fifty-two research participants were imaged at 3T using the 3D-EPI sequence acquired at different CAIPIRINHA acceleration factors (R = 2, 3, and 4) and denoised using a dedicated denoising convolutional neural network (DnCNN). Quantitative assessment of the accelerated T2*w 3D-EPI scans, before and after denoising, was performed using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and tissue contrasts. A neuroradiologist separately assessed image quality in a blinded manner using predetermined scoring criteria. T2*w 3D-EPI with CAIPIRINHA acceleration enabled fast submillimeter isotropic (0.65 mm) imaging of the entire brain with scan times ranging between 3 min 22 s (R = 2) down to 1 min 56 s (R = 4). Even for the fastest scan (R = 4), accelerated T2*w 3D-EPI images denoised with DnCNN exhibited superior PSNR (3 dB increase), SSIM (13% increase) and lesion-to-vein tissue contrast (9% increase) compared to the non-denoised images. The 3D-EPI sequence combined with CAIPIRINHA and deep learning denoising enables fast submillimeter whole-brain T2*w imaging at 3T.
Spin density-weighted (SDW) and inversion recovery (IR) 23Na MRI provide different sodium contrasts with complementary information. Therefore, the aim was to develop a time-efficient sequence scheme capable of providing both contrasts by acquiring SDW and IR 23Na MRI data within a single sequence without additional measurement time. In the developed pulse sequence, the conventional 180° inversion pulse was divided into two successive 90° pulses, with an additional readout inserted between them. The sequence was initially evaluated in simulations and phantom measurements. Afterwards, it was applied in calf muscle measurements of six healthy volunteers to assess its practical feasibility and signal characteristics. All 23Na measurements were acquired at 7 T and compared to standard SDW and IR sequences. The interleaved SDW/IR sequence yielded image quality and contrast comparable to standard SDW and IR sequences. Phantom experiments showed slightly higher remaining T1-weighting in the interleaved sequence compared with standard SDW, resulting in only minor changes in image contrast. Fluid suppression was effective across all IR approaches, though efficiency decreased modestly with longer effective inversion pulse lengths. Spin dynamic simulations for IR 23Na MRI indicated slightly increased sensitivity to fluid suppression artifacts arising from B0 inhomogeneities in the interleaved sequence. In vivo calf muscle measurements did not reveal significant differences in the mean muscle signal between the interleaved SDW/IR sequence and the corresponding standard sequences. The proposed interleaved sequence enables simultaneous acquisition of 23Na SDW and IR contrasts, thereby reducing overall scan time and eliminating the need for co-registration.
To develop a 3D reduced FOV sequence for combined ADC and T2 mapping in the prostate in a single scan. A 3D ADC and T2 mapping reduced FOV acquisition is enabled using T2 and diffusion preparation modules with slab-selective tip-down pulses and magnitude stabilizer gradients. Imaging shots are followed by 2D phase navigators for phase correction of diffusion-prepared shots and dummy RF pulses for T2-prepared shots, as well as a constant delay to maintain the steady state of the longitudinal magnetization. Mono-exponential fitting is used for ADC and T2 mapping. T2 mapping accuracy was assessed in a phantom against a single-echo spin-echo reference. In vivo, the sequence was compared in the prostate of 11 healthy volunteers against a multi-echo spin-echo acquisition, as well as standard and reduced FOV (rFOV) single-shot EPI (ssEPI). In the phantom experiments, T2 values were not significantly different from the reference single-echo spin-echo measurements. ADC values in vivo were not significantly different from those obtained using ssEPI. T2 values in vivo showed a mean bias of -21.7 ms compared to those obtained using multi-echo spin-echo with the first echo discarded in the fitting. The proposed sequence achieved similar overall image quality scores to those of ssEPI and rFOV ssEPI, improved perceived distortion scores, but lower perceived SNR scores. The proposed approach enables 3D reduced FOV, single-scan ADC, and T2 mapping with mono-exponential fitting in the prostate.
Noninvasive assessment of hepatic oxygenation is relevant for evaluating liver both in acute and long-term liver pathologies. However, existing methods are limited to single-slice acquisitions and require breath-holding. This study aims to develop and evaluate a motion-robust technique for whole-liver oxygen extraction fraction (OEF) mapping using a radial gradient-echo sampling of spin-echo (rGESSE) sequence in combination with an artificial neural network (ANN) for quantification. Seven healthy volunteers were scanned using a 1.5 T MRI system with an 18-channel body coil. A multi-slice rGESSE sequence with radial sampling was employed under free-breathing conditions to acquire volumetric liver data. Signal processing included bias-field correction and 3D median filtering. OEF, deoxygenated blood volume (DBV), and transverse relaxation rate (R2) were estimated using a trained feedforward ANN based on simulated qBOLD signal models. Whole-liver OEF maps were successfully obtained in all volunteers under free-breathing. Representative parameter maps showed consistent spatial patterns and anatomical correspondence. The mean hepatic OEF across subjects was 55.75% ± 6.88%, and the mean DBV was 0.568 ± 0.039. Comparison with literature values suggested a systematic overestimation, likely arising from a combination of model approximations, residual B0 inhomogeneity, motion effects, and acquisition-specific sensitivities. ANN-based fitting outperformed standard least-squares regression in terms of stability and artifact suppression. This study demonstrates the feasibility of using rGESSE combined with ANN analysis for free-breathing, whole-liver OEF mapping. The proposed approach allows for noninvasive, volumetric hepatic oxygenation assessment with improved motion robustness, offering potential for clinical application in liver disease diagnosis and monitoring.
In the spinal cord (SC), multi-echo gradient echo (ME-GRE) increases gray (GM) and white matter (WM) contrast and improves sensitivity to lesions in people with multiple sclerosis (pwMS). However, SC ME-GRE is susceptible to breathing-induced field fluctuations, causing ghosting artifacts and signal loss. Recent work introduced a 1D phase navigator following the last echo to measure field variations; however, susceptibility to phase wrapping increases at longer echo times. We propose a 1D phase navigator preceding the first echo, reducing phase accumulation and eliminating the need for respiratory monitoring. ME-GRE data covering the lower (T9-T12 vertebrae) and upper (T4-T8 vertebrae) thoracic SC were acquired in 20 healthy volunteers and 3 pwMS at 3T. Standard and navigator-corrected images were acquired in the same acquisition. To evaluate image quality, WM and GM signal-to-noise ratio (SNR), WM/GM contrast-to-noise ratio (CNR), and background ghosting signals were measured and compared between the two reconstructions. Both were blindly assessed for artifacts, structural delineation, and diagnostic confidence in pwMS. Navigator correction significantly increased GM and WM SNR and CNR, reduced posterior ghosting across both thoracic regions, and significantly reduced artifacts while increasing structural delineation. Preliminary evaluation in three pwMS showed consistent improvements in artifact mitigation, structural delineation, and lesion conspicuity with navigator correction, providing proof-of-concept for potential clinical application. A 1D navigator prior to the first echo reduces ghosting and improves thoracic SC image quality without respiratory monitoring. This approach could improve the diagnostic value and enhance the reliability of thoracic SC ME-GRE.
An MRI scanner was designed and built to encode k $$ k $$ -space points in the spin echoes occurring between 18 0 ∘ $$ 18{0}^{\circ } $$ pulses in an echo train by using blipped B 0 $$ {B}_0 $$ gradient pulses applied just prior to those spin echoes. The proposed MRI scanner is a modification of an original design that used TRansmit Array Spatial Encoding (TRASE), a built-in B 0 $$ {B}_0 $$ gradient, and a thin RF coil defining a single axial slice. However, for TRASE, the thin profile coil needed a spatially uniform B 1 $$ {B}_1 $$ phase, and this proved difficult to achieve. We therefore replaced the RF encoding coil with a low-power B 0 $$ {B}_0 $$ gradient coil. This removed the need for a uniform spatial B 1 $$ {B}_1 $$ phase for the single slice RF coil to achieve Fourier image encoding. The pulse sequence then sent all 18 0 ∘ $$ 18{0}^{\circ } $$ transmit RF pulses through the one remaining single slice RF coil with B 0 $$ {B}_0 $$ gradient pulses applied between the RF pulses. The spin echoes between the RF pulses were received and used to define the k $$ k $$ -space points in one transverse direction. Image encoding in the perpendicular transverse direction was accomplished via frequency encoding within the built-in transverse B 0 $$ {B}_0 $$ gradient. Mineral oil phantoms were used to establish a proof-of-concept for the blipped-gradient imaging approach. The phantom images give a proof-of-concept level verification of the blipped-gradient MRI design. With work to address engineering imperfections, a blipped-gradient approach could be used to achieve the extreme Size, Weight, and Power (SWaP) requirements for a spaceworthy MRI.
Conventional BOLD-fMRI relies on hemodynamic responses that are temporally and spatially indirect markers of neural activity. Developing alternative contrasts, sensitive to neuroelectrical phenomena, is a critical challenge in brain imaging. Spin-lock (SL) fMRI has shown promise in phantom studies for detecting magnetic field changes associated with neuronal activity, but its in-vivo sensitivity and practicality remain unclear. This study evaluated whether SL contrast can effectively detect and localize human neuronal activation, benchmarked against complementary functional modalities, magnetoencephalography (MEG) and 3T BOLD-fMRI, to assess the sensitivity of MR-based neuronal current imaging. Thirteen healthy young volunteers underwent SL-based imaging during 8 Hz visual stimulation, along with BOLD and MEG acquisitions. Subjects viewed quadrant-checkerboard stimuli to elicit localized cortical responses. Two balanced SL contrast mechanisms, rotary excitation (REX) and stimulus-induced rotary saturation (SIRS), were employed. Postprocessing targeted stimulus-locked signal fluctuations using a regression-filtering-rectification strategy. Phantom experiments tested sensitivity and analysis pipeline performance. MEG revealed robust stimulus-locked responses in the occipital cortex, with estimated local magnetic field amplitudes of ~0.07 nT. Conventional BOLD-fMRI confirmed reliable hemodynamic activation. In contrast, neither balanced REX nor balanced SIRS produced consistent stimulus-related activation in vivo. Phantom experiments subsequently yielded detection thresholds of 0.2 nT for REX and 0.6 nT for SIRS, exceeding the MEG-estimated physiological field amplitudes. Under the present experimental conditions, the tested spin-lock fMRI implementations did not achieve sufficient sensitivity for reliable in vivo detection of neuronal magnetic fields at 3T. Phantom and MEG-based estimates indicate that physiological field amplitudes in the visual cortex lie below current detection limits. These findings establish quantitative constraints on direct neuronal current imaging with MRI and provide a benchmark for future methodological developments aimed at bridging electrophysiology and functional MRI.
Preoperative bone mineral density (BMD) assessment is essential for cervical spine surgery, yet dual-energy X-ray absorptiometry (DXA) is not routinely performed. Opportunistic imaging metrics such as computed tomography-derived Hounsfield units (HU) and magnetic resonance imaging-based vertebral bone quality (VBQ) scores have been proposed as surrogates, but their comparative diagnostic value in the cervical spine remains unclear. In this retrospective single-center study, 154 patients aged ≥50 years undergoing cervical spine surgery for degenerative diseases were included. All patients underwent cervical CT, cervical MRI, DXA, and bone turnover marker testing within 3 months preoperatively. HU values at C2-C7 were measured on sagittal CT images, and VBQ scores were calculated from T1-weighted MRI. Correlations between imaging parameters and DXA T-scores (lumbar spine, femoral neck, and total hip) were analyzed. Receiver operating characteristic analysis evaluated diagnostic performance for reduced BMD (T-score <-1.0), including sex-specific subgroups. Mean cervical HU values decreased progressively from C4 to C7 and were significantly lower in patients with reduced BMD (all p < .001). C2 HU showed the strongest correlations with DXA T-scores (r = 0.528-0.671). C2 and C3 HU demonstrated numerically higher discriminative ability for identifying reduced BMD (area under the curve [AUC] = 0.737 and 0.741) compared with VBQ (AUC = 0.638) and Goutallier grade (AUC = 0.623). Performance of C2 HU was consistent across sexes. HU values showed weak correlations with β-cross-linked telopeptide of type I collagen and osteocalcin and no association with 25-hydroxyvitamin D or procollagen type I N-terminal propeptide. Upper cervical HU measurements, particularly at C2 and C3, demonstrated favorable discriminative performance for identifying reduced BMD and may serve as a practical opportunistic screening tool in patients undergoing cervical spine surgery.
Motion artifacts remain a major challenge in applying multi-shot 2D imaging to motion prone patient populations. Through-plane motion is especially problematic, where the lack of encoding cannot be easily recovered, even using deep-learning (DL)-regularized reconstruction. We demonstrate the benefits of combining prospective and retrospective motion correction, where the Scout Accelerated Motion Estimation and Reduction (SAMER) technique is utilized for on-the-fly motion estimation with field-of-view (FoV) updates along with retrospective correction of potential residual motion. Four prospective motion correction (pMoCo) strategies were implemented within a custom 2D turbo-spin-echo (TSE) SAMER enabled sequence. They were evaluated in vivo across representative subject motion, with associated simulations to characterize artifacts and the correction performance. In addition, motion trajectories measured during inpatient clinical exams were used to further demonstrate the robustness of the combined motion correction approach. Prospectively applying FoV updates significantly improved the image quality of SAMER reconstructions. Simulated artifact patterns were shown to closely match those observed in vivo, and across 274 simulations using clinical motion trajectories, the combined approach reduced NRMSE in 90% of moderate-to-severe motion cases and significantly decreased the overall reconstruction error. Utilizing on-the-fly SAMER motion estimates, a combined prospective and retrospective motion correction approach was demonstrated for 2D TSE imaging. The proposed method improved image quality in several representative in vivo motion experiments and across simulations of a wide range of clinical inpatient motion conditions. In addition, simulated artifact patterns were shown to closely match those observed in vivo. This capability should enable on-the-fly prediction of motion artifacts for efficient/intelligent acquisition strategies for the most challenging motion scenarios.
To implement and evaluate chemical exchange saturation transfer (CEST) at ultra-high field on a whole-body 11.7 T scanner using parallel transmission (pTx). Tailored pTx CEST saturation pulses were optimized to achieve a spatially uniform target B1 + while satisfying SAR and power constraints either defined by conservatively defined radio-frequency (RF) power limits or by a virtual observation point (VOP)-based framework. The resulting pulse performance at 11.7 T was assessed through numerical simulations and compared against corresponding designs at 7 T. The proposed saturation pulses were integrated into a two-dimensional multi-slice EPI sequence. Experimental testing of the sequence was performed in vitro using metabolite-doped phantoms and in vivo in four healthy volunteers. The pTx pulse designs optimized under VOP-based SAR constraints at 11.7 T achieved comparable performance to 7 T designs despite higher intrinsic B1 + inhomogeneity. In vitro experiments validated the sequence implementation and confirmed detection of the targeted metabolites. In vivo experiments demonstrated acceptable saturation homogeneity (nRMSE ≤ 12.1%) in a slab of thickness 1 cm, with only minimal additional improvement achieved through B1 + post-processing correction. Across the cohort, robust gray matter-white matter contrast was consistently observed for multiple CEST contrasts. VOP-based SAR-constrained pTx CEST saturation pulse design enabled robust and reproducible contrast at 11.7 T by effectively mitigating B1 + inhomogeneities while remaining within safety limits, facilitating translation to future in vivo clinical research applications.
Pulmonary hypertension is related to a 1-year mortality ranging from < 5% to > 20% in low- and high-risk groups. Right heart catheterization is the diagnostic gold standard, associated with a 1.1% rate of severe adverse events, posing the need for a non-invasive diagnostic tool. This study aims to delineate the effects of controlled changes on pulmonary hemodynamics by quantifying corresponding changes in hyperpolarized 129Xe MRSI dissolved-phase parameters, including RBC oscillations. Seven Danish Landrace pigs underwent hyperpolarized 129Xe dissolved-phase 3D MRSI with interleaved MRS under baseline conditions and three vasoactive interventions, monitored with invasive catheters or phase-contrast MRI. Results were evaluated using one-way ANOVA or mixed-effects analysis followed by Dunnett's multiple comparisons test. Hypoxia increased mPAP (p = 0.038), reduced RBC:M (-31%, p = 0.007) and RBC:G (-34%, p = 0.009), while increasing RBC chemical shift (0.57%, p = 0.020). Adenosine and dobutamine increased RBC:M (22%, p = 0.003) and RBC:G (30%, p = 0.009), while decreasing RBC chemical shift (-0.38%, p = 0.11). Linear regression revealed an inverse correlation between mPAP and RBC oscillation amplitude (R2 = 0.158, p = 0.04), RBC:M (R2 = 0.25, p = 0.05), and RBC:G (R2 = 0.22, p = 0.01). This study demonstrates that dissolved-phase hyperpolarized 129Xe MRSI parameters are sensitive to hemodynamic changes, exhibiting consistent inverse relationships with invasively measured mPAP. By leveraging concurrent invasive reference measurements, these findings provide physiological validation that dissolved-phase 129Xe metrics reflect pulmonary vascular pressure, supporting their further development as a non-invasive adjunct to right heart catheterization.
To accelerate MRI further, rapid B0 field modulations can be applied during oversampled readout to capture additional physical information, as in Wave-CAIPI/FRONSAC/local B0 coils modulation techniques. These methods, however, turn the Fourier readout into a non-Fourier-encoded dimension that cannot be reconstructed by FFT, posing significant reconstruction challenges especially in compressed-sensing or neural-network frameworks. Because the rapid B0 modulations still vary slowly relative to the oversampled ADC dwell time, we exploit this encoding redundancy by compressing k-space patch-by-patch across subregions, each of which is jointly encoded by a distinct subset of B0 and RF (receive) spatial encoding functions. For each subset, a compression matrix is computed once and reused to compress all patches encoded by the same B0-RF spatial modulations. This can be implemented by feeding subsets of B0 and RF spatial encoding maps into an adapted conventional RF array compression algorithm, mimicking an expanded set of virtual receiver channels. This approach was evaluated on human brain scans at 9.4 T/3 T. The proposed group-patch joint compression achieves substantially higher compression factors than conventional RF-only compression, while minimally compromising encoding efficiency. Typically, joint compression factors of 11×-20× led to negligible encoding loss, dramatically reducing reconstruction time and peak memory usage. For example, compressed-sensing reconstruction took 1.4-5.1 s/2D slice, 177 s-10.1 min/3D volume on a high-memory CPU node. Given joint encoding of dynamic B0 and static RF fields, compressing multidimensional k-space patches in separate groups outperforms compressing RF receivers alone. This substantially mitigates a fundamental computational bottleneck when combining rapid B0 and RF-receiver modulations.