Accelerating quantitative MR imaging with the incorporation of B1 compensation using deep learning

Quantitative magnetic resonance imaging (MRI) attracts attention due to its support to quantitative image analysis and data driven medicine. However, the application of quantitative MRI is severely limited by the long data acquisition time required by repetitive image acquisition and measurement of field map. Inspired by recent development of artificial intelligence, we propose a deep learning strategy to accelerate the acquisition of quantitative MRI, where every quantitative T₁ map is derived from two highly undersampled variable-contrast images with radiofrequency field inhomogeneity automatically compensated. In a multi-step framework, variable-contrast images are first jointly reconstructed from incoherently undersampled images using convolutional neural networks; then T₁ map and B₁ map are predicted from reconstructed images employing deep learning. Thus, the acceleration includes undersampling in every input image, a reduction in the number of variable contrast images, as well as a waiver of B₁ map measurement. The strategy is validated in T₁ mapping of cartilage. Acquired with a consistent imaging protocol, 1224 image sets from 51 subjects are used for the training of the prediction models, and 288 image sets from 12 subjects are used for testing. High degree of acceleration is achieved with image fidelity well maintained. The proposed method can be broadly applied to quantify other tissue properties (e.g. T₂, T_1ρ) as well.     

Targeted radiofrequency field mapping using 3D reduced field-of-view-catalyzed double-angle method

Purpose

The aim of this study was to develop a targeted volumetric radiofrequency field (B₁⁺) mapping technique to provide region-of-interest B₁⁺ information.

Materials and Methods

Targeted B₁⁺ maps were acquired using three-dimensional (3D) reduced field-of-view (FOV) inner-volume turbo spin echo-catalyzed double-angle method (DAM). Targeted B₁⁺ maps were compared with full-FOV B₁⁺ maps acquired using 3D catalyzed DAM in a phantom and in the brain of a healthy volunteer. In addition, targeted volumetric abdomeninal B₁⁺ mapping was demonstrated in the abdomen of another healthy volunteer.

Results

The targeted reduced-FOV images demonstrated no aliasing artifacts in all experiments. Close match between targeted B₁⁺ map and reference full-FOV B₁⁺ map in the same region was observed, with percentage root-mean-squared error <0.4% in the phantom and <0.8% in the healthy volunteer brain. The abdominal B₁⁺ maps showed small B₁⁺ variation in the kidneys and liver from the healthy volunteer.

Conclusion

The proposed 3D reduced-FOV catalyzed DAM provides a rapid, simple and accurate method for targeted volumetric B₁⁺ mapping and can be easily implemented for applications related to radiofrequency field mapping in small targeted regions.     

Deriving new soft tissue contrasts from conventional MR images using deep learning

Versatile soft tissue contrast in magnetic resonance imaging is a unique advantage of the imaging modality. However, the versatility is not fully exploited. In this study, we propose a deep learning-based strategy to derive more soft tissue contrasts from conventional MR images obtained in standard clinical MRI. Two types of experiments are performed. First, MR images corresponding to different pulse sequences are predicted from one or more images already acquired. As an example, we predict T_1ρ weighted knee image from T₂ weighted image and/or T₁ weighted image. Furthermore, we estimate images corresponding to alternative imaging parameter values. In a representative case, variable flip angle images are predicted from a single T₁ weighted image, whose accuracy is further validated in quantitative T₁ map subsequently derived. To accomplish these tasks, images are retrospectively collected from 56 subjects, and self-attention convolutional neural network models are trained using 1104 knee images from 46 subjects and tested using 240 images from 10 other subjects. High accuracy has been achieved in resultant qualitative images as well as quantitative T₁ maps. The proposed deep learning method can be broadly applied to obtain more versatile soft tissue contrasts without additional scans or used to normalize MR data that were inconsistently acquired for quantitative analysis.     

Complementary use of T2-weighted and postcontrast T1- and T21-weighted imaging to distinguish sites of reversible and irreversible brain damage in focal ischemic lesions in the rat brain

The evolution of a photochemically induced cortical infarct was monitored using T₂-, postcontrast (GdDOTA) T₁-, and postcontrast (DyDTPA-BMA) T₂¹-weighted NMR imaging techniques. Data acquired with these different NMR imaging types were compared, both qualitatively and quantitatively. The T₂¹-weighted NMR images after sprodiamide injection (DyDTPA-BMA) were perfusion-weighted images that allowed the differentiation between several infarct-related areas in terms of different degrees of perfusion deficiency. No quantitative information on cerebral blood flow (CBF) was obtained. A clear distinction was made between areas with a complete lack of CBF located in the core of the lesion and temporary CBF insufficiencies in the rim surrounding this core. Concomitant observations on T₂-weighted and postcontrast T₁-weighted images revealed the same temporary rim characterized by an increased water content, and an intact blood-brain barrier (BBB), as well as by reduced perfusion. This rim appeared within the first hours after infarct induction, reached a maximum 24 h later, and lasted between 3–5 days, when its size gradually decreased until complete disappearance. These observations suggest the existence of an area at risk. Only on postcontrast T₁-weighted images, the core of the lesion remained visible during the whole experimental period (10 days) and reflected in all likelihood the irreversibly damaged ischemic central core. The combined application of different NMR imaging techniques when studying focal cerebral infarctions in the rat brain allowed us to distinguish, in terms of NMR characteristics, zones of reversible from irreversible brain damage and to estimate the severity of the damage. This might offer an appropriate experimental setup for the screening of cerebroprotective compounds.     

B(1)(+)/actual flip angle and reception sensitivity mapping methods: simulation and comparison

Knowledge of the spatial distribution of transmission field B₁⁺ and reception sensitivity maps is important in high-field (≥3 T) human magnetic resonance (MR) imaging for several reasons: these include post-acquisition correction of intensity inhomogeneities, which may affect the quality of images; modeling and design of radiofrequency (RF) coils and pulses; validating theoretical models for electromagnetic field calculations; testing the compatibility with MR environment of biomedical implants. Moreover, inhomogeneities in the RF field are an essential source of error for quantitative MR spectroscopy. Recent studies have also shown that B₁⁺ and reception sensitivity maps can be used for direct calculation of tissue electrical parameters and for estimating the local specific absorption rate (SAR) in vivo.Several B₁⁺ mapping techniques have been introduced in the past few years based on actual flip angle (FA) mapping, but, to date, none has emerged as a standard. For reception sensitivity calculation, the signal intensity equation can be used where the nominal FA distribution must be replaced with the actual FA distribution calculated by one of the B₁⁺ mapping techniques.This study introduces a quantitative comparison between two known methods for B₁⁺/actual FA and reception sensitivity mapping: the double-angle method (DAM) and the fitting (FIT) method. Experimental data obtained using DAM and FIT methods are also compared with numerical simulation results.     

Channel-combination method for phase-based |B1+| mapping techniques

PurposeThe aim of this study was to propose a channel combination method for |B₁⁺| mapping methods using phase difference to reconstruct |B₁⁺| map.Theory and methodsPhase-based |B₁⁺| mapping methods commonly consider the phase difference of two scans to measure |B₁⁺|. Multiple receiver coils acquire a number of images and the phase difference at each channel is theoretically the same in the absence of noise. Affected by noise, phase difference is approximately governed by Gaussian distribution. Considering data from all channels as samples, estimation can be achieved by maximum likelihood method. With this method, all phase differences at each channel are combined into one. In this study, the proposed method is applied with Bloch-Siegert shift |B₁⁺| mapping method. Simulations are performed to illustrate the phase difference distribution and demonstrate the feasibility and facility of the proposed method. Phantom and vivo experiments are carried out at 1.5 T scanner equipped with 8-channel receiver coil. In all experiments, the proposed method is compared with weighted averaging (WA) method.ResultsSimulations revealed appropriateness of approximating the distribution of phase difference to Gaussian distribution. Compared with WA method, the proposed method reduces errors of |B₁⁺| calculation. Phantom and vivo experiments provide further validation.ConclusionConsidering phase noise distribution, the proposed method achieves channel combination by finding the estimation from data acquired by multiple receivers coil. The proposed method reduces |B₁⁺| reconstruction errors caused by noise.     

Combined17O/1H MRI study in a whole-body scanner

A simple method of obtaining consecutive¹H and natural-abundance¹⁷O images is described with a scanner’s original body resonator (for¹H) and a homemade linear birdcage (for¹⁷O). Two kinds of experiments were performed to test the method. In the first experiment, a proton image of the phantom was acquired with a whole-body resonator. In the second experiment, the phantom was inserted into an oxygen birdcage resonator and imaged again with a whole-body resonator. The intensities of images resulting from the experiments were analyzed. Although theB ₁ field homogeneity is disturbed, the proton images acquired with a whole-body resonator when the oxygen resonator is present are of acceptable quality for use in the combined¹⁷O/¹H imaging.     

TriTone: a radiofrequency field (B1)-insensitive T1 estimator for MRI at high magnetic fields

Fast, high-resolution, longitudinal relaxation time (T1) mapping is invaluable in clinical and research applications. It has been shown that two spoiled gradient recalled echo (SPGR) images acquired in steady state with variable flip angles is an attractive alternative to the multi-image sets previously acquired with inversion or saturation recovery. The known sensitivity of the two-point method to transmit radiofrequency field (B1) inhomogeneity exacerbated at 3 T and above, however, mandates its combination with an additional, time-consuming and possibly specific-absorption-rate-intensive B1 measurement, preventing direct migration of the method to these fields. To address this, we introduce a method designed to be free of systematic errors caused by B1 inhomogeneity in which the value of T1 is extracted from three SPGR images acquired with echo planar imaging (EPI) readout. The precision of the T1 maps produced is found to be comparable to the two-point method, while the accuracy is greatly improved in the same time and spatial resolution. A welcome byproduct of the method is a map of B1 that can be used to correct other acquisitions in the same session. Tables of the optimal acquisition protocols are provided for several total imaging times.     

Reducing bias in DREAM flip angle mapping in human brain at 7T by multiple preparation flip angles

DREAM (Dual Refocusing Echo Acquisition Mode) is an ultra-fast multi-slice B₁⁺-mapping technique based on the single-shot STEAM sequence. To study systematic errors at high actual flip angles (FA) and low SNR, DREAM B₁⁺ maps at 3.75×3.75×3.50 mm³ resolution were acquired at 7T in phantoms and human brain in vivo with nominal FAs between 20° and 90° for the two STEAM preparation pulses. Predicted B₁⁺ estimates were underestimated at actual FAs above 50° while noise was prominent below 20°. With a reliable interval of the actual FA between 20° and 50° identified, a B₁⁺ range of 33% - 200% of nominal FA is covered by varying the nominal preparation angle through 25°, 40°, and 60°. Individual B₁⁺ maps are thresholded according to the identified interval and combined into a single map. We demonstrate the benefit of the combined low-noise, low-bias B₁⁺ maps for dual flip angle T₁-mapping.     

Magnetic Resonance Imaging of Gases: A Single-Point Ramped Imaging with T1 Enhancement (SPRITE) Study

A pure phase-encoding MRI technique, single-point ramped imaging withT₁enhancement, SPRITE, is introduced for the purpose of gas phase imaging. The technique utilizes broadband RF pulses and stepped phase encode gradients to produce images, substantially free of artifacts, which are sensitive to the gasT₁andT^*:₂relaxation times. Images may be acquired from gas phase species with transverse relaxation times substantially less than 1 ms. Methane gas images,¹H, were acquired in a phantom study. Sulfur hexafluoride,¹⁹F, images were acquired from a gas-filled porous coral sample. High porosity regions of the coral are observed in both the MRI image and an X-ray image. Sensitivity and resolution effects due to signal modulation during the time-efficient acquisition are discussed. A method to increase the image sensitivity is discussed, and the predicted improvement is shown through 1D images of the methane gas phantom.     

Fast and high-resolution quantitative mapping of tissue water content with full brain coverage for clinically-driven studies

An efficient method for obtaining longitudinal relaxation time (T1) maps is based on acquiring two spoiled gradient recalled echo (SPGR) images in steady states with different flip angles, which has also been extended, with additional acquisitions, to obtain a tissue water content (M0) map. Several factors, including inhomogeneities of the radio-frequency (RF) fields and low signal-to-noise ratios may negatively affect the accuracy of this method and produce systematic errors in T1 and M0 estimations. Thus far, these limitations have been addressed by using additional measurements and applying suitable corrections; however, the concomitant increase in scan time is undesirable for clinical studies. In this note, a modified dual-acquisition SPGR method based on an optimization of the sequence formulism is presented for good and reliable M0 mapping with an isotropic spatial resolution of 1 × 1 × 1 mm³ that covers the entire human brain in 6:30 min. A combined RF transmit/receive map is estimated from one of the SPGR scans and the optimal flip angles for M0 map are found analytically. The method was successfully evaluated in eight healthy subjects producing mean M0 values of 69.8% (in white matter) and 80.1% (in gray matter) that are in good agreement with those found in the literature and with high reproducibility. The mean value of the resultant voxel-based coefficients-of-variation was 3.6%.     

Fractional order vs. exponential fitting in UTE MR imaging of the patellar tendon

PurposeQuantification of the T₂¹ relaxation time constant is relevant in various magnetic resonance imaging applications. Mono- or bi-exponential models are typically used to determine these parameters. However, in case of complex, heterogeneous tissues these models could lead to inaccurate results. We compared a model, provided by the fractional-order extension of the Bloch equation with the conventional models.MethodsAxial 3D ultra-short echo time (UTE) scans were acquired using a 3.0 T MRI and a 16-channel surface coil. After image registration, voxel-wise T₂¹ was quantified with mono-exponential, bi-exponential and fractional-order fitting. We evaluated all three models repeatability and the bias of their derived parameters by fitting at various noise levels. To investigate the effect of the SNR for the different models, a Monte-Carlo experiment with 1000 repeats was performed for different noise levels for one subject. For a cross-sectional investigation, we used the mean fitted values of the ROIs in five volunteers.ResultsComparing the mono-exponential and the fractional order T₂¹ maps, the fractional order fitting method yielded enhanced contrast and an improved delineation of the different tissues. In the case of the bi-exponential method, the long T₂¹ component map demonstrated the anatomy clearly with high contrast. Simulations showed a nonzero bias of the parameters for all three mathematical models. ROI based fitting showed that the T₂¹ values were different depending on the applied method, and they differed most for the patellar tendon in all subjects.ConclusionsIn high SNR cases, the fractional order and bi-exponential models are both performing well with low bias. However, in all observed cases, one of the bi-exponential components has high standard deviation in T₂¹. The bi-exponential model is suitable for T₂¹ mapping, but we recommend using the fractional order model for cases of low SNR.     

A statistical fMRI model for differential T2* contrast incorporating T1 and T2* of gray matter

Relaxation parameter estimation and brain activation detection are two main areas of study in magnetic resonance imaging (MRI) and functional magnetic resonance imaging (fMRI). Relaxation parameters can be used to distinguish voxels containing different types of tissue whereas activation determines voxels that are associated with neuronal activity. In fMRI, the standard practice has been to discard the first scans to avoid magnetic saturation effects. However, these first images have important information on the MR relaxivities for the type of tissue contained in voxels, which could provide pathological tissue discrimination. It is also well-known that the voxels located in gray matter (GM) contain neurons that are to be active while the subject is performing a task. As such, GM MR relaxivities can be incorporated into a statistical model in order to better detect brain activation. Moreover, although the MR magnetization physically depends on tissue and imaging parameters in a nonlinear fashion, a linear model is what is conventionally used in fMRI activation studies. In this study, we develop a statistical fMRI model for Differential T₂^? ConTrast Incorporating T₁ and T₂^? of GM, so-called DeTeCT-ING Model, that considers the physical magnetization equation to model MR magnetization; uses complex-valued time courses to estimate T₁ and T₂^? for each voxel; then incorporates gray matter MR relaxivities into the statistical model in order to better detect brain activation, all from a single pulse sequence by utilizing the first scans.     

Correction of errors caused by imperfect inversion pulses in MR imaging measurement of T1 relaxation times

Spin-lattice (T₁) relaxation times were measured by an inversion-recovery magnetic resonance imaging method with a slice-selective inversion pulse (SIP), a non-selective rectangular inversion pulse (RIP), or a B₁-insensitive adiabatic inversion pulse (AIP). Data analysis either assumed perfect inversion (two-parameter fit) or allowed for imperfect inversion (three-parameter fit). Imperfect inversion pulses caused low T₁ values in phantoms with a two-parameter fit, while three-parameter T₁ estimates were accurate over the range 430–2670 ms. A difference of ∼10% between two-parameter and three-parameter T₁ values in normal human brain tissue was attributed to B₁ inhomogeneity with the slice-selective inversion pulse and rectangular inversion pulse, to the slice profile with the slice-selective inversion pulse, and to T₂ effects for the adiabatic inversion pulse. Any T₁ method that relies on accurate flip angles may have a significant systematic error in vivo. Phantom accuracy does not ensure accuracy in vivo, because phantoms may have a more homogeneous B₁ field and a longer T₂ than do biological samples.     

Parallel-plate RF resonator imaging of chemical shift resolved capillary flow

Magnetic resonance imaging has been introduced to study flow in microchannels using pure phase spatial encoding with a microfabricated parallel-plate nuclear magnetic resonance (NMR) probe. The NMR probe and pure phase spatial encoding enhance the sensitivity and resolution of the measurement. In this paper, ¹H NMR spectra and images were acquired at 100 MHz. The B₁ magnetic field is homogeneous and the signal-to-noise ratio of 30 μl doped water for a single scan is 8×10⁴. The high sensitivity of the probe enables velocity mapping of the fluids in the micro-channel with a spatial resolution of 13×13 μm. The parallel-plate probe with pure phase encoding permits the acquisition of NMR spectra; therefore, chemical shift resolved velocity mapping was also undertaken. Results are presented which show separate velocity maps for water and methanol flowing through a straight circular micro-channel. Finally, future performance of these techniques for the study of microfluidics is extrapolated and discussed.     

Rapid high-resolution volumetric T1 mapping using a highly accelerated stack-of-stars Look Locker technique

PurposeTo develop a fast volumetric T₁ mapping technique.Materials and methodsA stack-of-stars (SOS) Look Locker technique based on the acquisition of undersampled radial data (>30× relative to Nyquist) and an efficient multi-slab excitation scheme is presented. A principal-component based reconstruction is used to reconstruct T₁ maps. Computer simulations were performed to determine the best choice of partitions per slab and degree of undersampling. The technique was validated in phantoms against reference T₁ values measured with a 2D Cartesian inversion-recovery spin-echo technique. The SOS Look Locker technique was tested in brain (n = 4) and prostate (n = 5). Brain T₁ mapping was carried out with and without k_z acceleration and results between the two approaches were compared. Prostate T₁ mapping was compared to standard techniques. A reproducibility study was conducted in brain and prostate. Statistical analyses were performed using linear regression and Bland Altman analysis.ResultsPhantom T₁ values showed excellent correlations between SOS Look Locker and the inversion-recovery spin-echo reference (r² = 0.9965; p < 0.0001) and between SOS Look Locker with slab-selective and non-slab selective inversion pulses (r² = 0.9999; p < 0.0001). In vivo results showed that full brain T₁ mapping (1 mm³) with k_z acceleration is achieved in 4 min 21 s. Full prostate T₁ mapping (0.9 × 0.9 × 4 mm³) is achieved in 2 min 43 s. T₁ values for brain and prostate were in agreement with literature values. A reproducibility study showed coefficients of variation in the range of 0.18–0.2% (brain) and 0.15–0.18% (prostate).ConclusionA rapid volumetric T₁ mapping technique was developed. The technique enables high-resolution T₁ mapping with adequate anatomical coverage in a clinically acceptable time.     

Quantitative T(1)(ρ) imaging using phase cycling for B0 and B1 field inhomogeneity compensation

T_1ρ imaging is useful in a number of clinical applications. T_1ρ preparation methods, however, are sensitive to non-uniformities of the B₀ magnetic field and the B₁ RF field. These common system imperfections can result in image artifacts and quantification errors in T_1ρ imaging. We report on a phase-cycling method which can eliminate B₁ RF inhomogeneity effects in T_1ρ imaging. This method does not only correct for image artifacts but also for T_2ρ contamination caused by B₁ RF inhomogeneity. The presence of B₀ magnetic field inhomogeneity can compromise the effectiveness of this method for B₁ RF inhomogeneity correction. We demonstrate that, by combining the spin-locking scheme reported by Dixon et al. (Myocardial suppression in vivo by spin locking with composite pulses. Magn Reson Med 1996; 36:90-94) with phase cycling, we can simultaneously correct B₀ magnetic field inhomogeneity effects and B₁ RF inhomogeneity effects in T_1ρ imaging. Phantom and in vivo data sets are used to demonstrate the proposed methods and to compare them with other existing T_1ρ preparation methods.     

Labeling of macrophages using bacterial magnetosomes and their characterization by magnetic resonance imaging

This work investigated macrophages labeled with magnetosomes for the possible detection of inflammations by MR molecular imaging. Pure magnetosomes and macrophages containing magnetosomes were analyzed using a clinical 1.5 T MR-scanner. Relaxivities of magnetosomes and relaxation rates of cells containing magnetosomes were determined. Peritonitis was induced in two mice. T₁, T₂ and T₂* weighted images were acquired following injection of the probes. Pure magnetosomes and labeled cells showed slight effects on T₁, but strong effects on T₂ and T₂* images. Labeled macrophages were located with magnetic resonance imaging (MRI) in the colon area, thus demonstrating the feasibility of the proposed approach.     

Clinical high-resolution mapping of the proteoglycan-bound water fraction in articular cartilage of the human knee joint

PurposeWe applied our recently introduced Bayesian analytic method to achieve clinically-feasible in-vivo mapping of the proteoglycan water fraction (PgWF) of human knee cartilage with improved spatial resolution and stability as compared to existing methods.Materials and methodsMulticomponent driven equilibrium single-pulse observation of T₁ and T₂ (mcDESPOT) datasets were acquired from the knees of two healthy young subjects and one older subject with previous knee injury. Each dataset was processed using Bayesian Monte Carlo (BMC) analysis incorporating a two-component tissue model. We assessed the performance and reproducibility of BMC and of the conventional analysis of stochastic region contraction (SRC) in the estimation of PgWF. Stability of the BMC analysis of PgWF was tested by comparing independent high-resolution (HR) datasets from each of the two young subjects.ResultsUnlike SRC, the BMC-derived maps from the two HR datasets were essentially identical. Furthermore, SRC maps showed substantial random variation in estimated PgWF, and mean values that differed from those obtained using BMC. In addition, PgWF maps derived from conventional low-resolution (LR) datasets exhibited partial volume and magnetic susceptibility effects. These artifacts were absent in HR PgWF images. Finally, our analysis showed regional variation in PgWF estimates, and substantially higher values in the younger subjects as compared to the older subject.ConclusionsBMC-mcDESPOT permits HR in-vivo mapping of PgWF in human knee cartilage in a clinically-feasible acquisition time. HR mapping reduces the impact of partial volume and magnetic susceptibility artifacts compared to LR mapping. Finally, BMC-mcDESPOT demonstrated excellent reproducibility in the determination of PgWF.     

Quantitative NMR microscopy on intact plants

Quantitative high resolution images on intact young maize plants were acquired by using magnetization-prepared NMR microscopy. Although the spatial resolution is low compared with that of light microscopy, the calculated spin density and T₁ maps exhibit contrasts that are in excellent agreement with photomicrographic images. The T₂ map gives image contrasts that are not visible in a usual light microscopic image. The diffusion images show an anisotropic behavior of the water self-diffusion coefficient in the vascular bundles, which can be understood by the cell morphology in this plant section. This work demonstrates that quantitative imaging on intact plant systems is possible and that long total acquisition times are no obstacle. Furthermore, the different single parameter maps give a better insight into the morphology of plants under in vivo conditions.