Magnetic Resonance Materials in Physics, Biology and Medicine最新文献

Objective: Quantifying myelin is essential for understanding demyelinating disease, yet myelin water fraction (MWF) varies across studies due to sequence choice, acquisition parameters, and field inhomogeneities. A major limitation is the lack of a phantom with a known structural reference. This study aimed to develop an electron microscopy (EM)-derived numerical MRI phantom representing structural tissue water fractions and to evaluate how acquisition and field conditions influence simulated MWF estimates. Here, simulated MWF refers to the short-T₂ component recovered from EM-derived structural fractions and should be distinguished from in-vivo biological MWF.

Materials and methods: EM images of CNS tissue were segmented into myelin, axons, and intra-/extracellular water (29%, 41%, 30%). Relaxation times were assigned from the literature, and simulated MWF was defined as the short-T₂ component estimated using non-negative least squares. Multi-echo spoiled gradient echo signals were simulated across TR, flip angle, and synthetic B₀/B₁ inhomogeneity.

Results: The phantom produced compartment-specific decay curves and enabled controlled evaluation of acquisition-dependent behavior. MWF increased with flip angle, decreased with longer TR, and showed systematic bias under B₀/B₁ variation. The framework further visualized mixing among compartments and sensitivity to relaxation- and field-driven changes.

Discussion: This tissue water fraction phantom provides a structural ground truth for reproducible evaluation of simulated MWF and supports optimization and future methodological standardization of quantitative MRI.

Objective: Modern computational MRI denoising approaches are often designed assuming fixed k-space coverage. This contrasts with earlier acquisition-design literature that leveraged k-space coverage modifications (e.g., reducing spatial resolution) to improve SNR. This work investigates whether the performance of modern computational denoising methods can be further enhanced by k-space coverage modifications.

Materials and methods: Using realistic simulations of noisy data, k-space coverage and averaging patterns were optimized for two advanced image denoising/reconstruction approaches: parallel imaging with total variation regularization and a U-Net neural network. For reference, comparisons against classical linear filtering/apodization methods were also performed. Performance was quantified using normalized root-mean-squared error (NRMSE) and structural similarity (SSIM) metrics.

Results: Advanced computational denoising methods can be substantially enhanced, both quantitatively and qualitatively, by reducing the spatial resolution of the acquisition to improve SNR. Indeed, even simple linear filtering/apodization with optimized k-space coverage can rival advanced methods using naive higher-resolution coverage.

Discussion: Classical acquisition design principles that allow spatial resolution to be traded for SNR enhancement are still very relevant for modern computational denoising techniques. However, the optimization of k-space coverage and denoising/reconstruction methods can also be somewhat confounded because the NRMSE and SSIM metrics have low sensitivity to spatial resolution.

Purpose: To facilitate ease of data compilation across diverse populations for training models to synthesize clinical contrast-weighted images from magnetic resonance fingerprinting.

Methods: We leverage a semi-supervised training framework using highly accelerated acquisitions of the target contrasts used as ground truths. We utilize complementary randomized data sampling masks across training subjects and contrasts for homogeneous learning in k-space, together with multi-task learning.

Results: Our experiments indicate that the proposed method achieves high-quality synthesis with networks trained on retrospectively and prospectively undersampled data of the contrast-weighted images, enabling undersampling up to 12-16 $\times$ .

Conclusions: The proposed method enables semi-supervised learning for synthesis from MRF with an end-to-end, ultra-fast training data acquisition protocol that is easier to obtain across a large population in clinical settings.

Objective: We explored the feasibility of our pipeline to make semi-automated estimations of the cervical canal area (CCaA), a proxy for spinal cord reserve in multiple sclerosis (MS), using conventional MRI sequences other than 3D T1-weighted images (T1WI), with which the pipeline was initially validated.

Materials and methods: Fifty-three patients with MS underwent sagittal brain 3DT1WI and cervical 2DT1WI, T2WI, and short tau inversion recovery (STIR). Semi-automated CCaA estimations were obtained from reconstructed axial images at the C2/C3 and C3/C4 levels. Agreement with manual segmentations was evaluated using the Dice coefficient (DC). Intraclass correlation coefficients (ICC) assessed consistency across sequences, and Spearman correlation tested associations with Expanded Disability Status Scale (EDSS).

Results: After quality control, the final cohort comprised 36 patients with MS at the C2/C3 level and 43 at C3/C4. CCaA estimation failed on 2DT1WI. Agreement with manual masks was stronger for T2WI (DC = 0.92 [0.89-0.93]) and STIR (DC = 0.90 [0.88-0.92]). The equivalence of CCaA across sequences was higher at C3/C4 than at C2/C3: ICC T2WI-3DT1WI was 0.67 (0.38-0.82) and 0.63 (0.26-0.82), while ICC STIR-3DT1WI was 0.80 (0.64-0.89) and 0.52 (0.22-0.70), respectively. At C3/C4, T2WI CCaA and EDSS were significantly correlated (rho -0.34, p 0.023).

Discussion: CCaA can be reliably estimated from 2DT2WI and STIR images. T2WI-derived CCaA was significantly associated with disability.

Introduction: The aim of this work was to investigate the effect of an air layer of different thicknesses on the design of high-permittivity materials (HPM) helmets for ultrahigh field (UHF) magnetic resonance imaging (MRI).

Method: We used a recently proposed analytical formulation of scattering from layered spheres to model an MRI experiment with a variable air layer between a homogenous human head and an HPM helmet. Such model expresses the fields as a superposition of progressive and regressive traveling waves by exploiting the theory of inhomogeneous transmission. Analytical results were compared with numerical simulations, in terms of changes in the radiofrequency (RF) magnetic induction field employing a surface and volume coil, to validate the proposed method.

Results: The presence of an air layer, due to differences in head size, results in a slight variation in the optimal permittivity value required to optimize the performance of the helmet, with a maximum relative variation of no more than 12%. This can be explained by the invariance of the impedance at the outer air-HPM interface, due to the high conductivity typical of biological tissues. In both cases, a clear increase in the magnetic induction field is observed, suggesting that the HPM design is robust to the introduction of a small dielectric insulating layer. Also, good agreement was found between the analytical and numerical results suggesting that the model could be employed to optimize the HPM also in real experiments, particularly when canonical geometries, such as cylindrical or spherical shapes, are employed to design the helmet.

Objective: To develop and evaluate a free-breathing 2D radial joint T₁-T₂ mapping technique for the kidneys at 3T, and to assess the impact of navigator gating parameters on mapping accuracy in a phantom and precision in healthy volunteers.

Methods: The PARMANav sequence (PArametric Radial MApping with Navigator gating) was implemented with 25 single-shot radial gradient-echo acquisitions with five magnetization preparations and lung-liver navigator gating to avoid through-plane motion. Images were reconstructed using region-optimized virtual coils and compressed sensing, followed by model-based registration. An acquisition-specific joint T₁-T₂ dictionary was generated using extended phase-graph simulations. T₁-T₂ accuracy was quantified in a phantom and T₁-T₂ precision was established in 10 healthy volunteers. Three patients were scanned to demonstrate clinical feasibility.

Results: In the phantom, PARMANav T₁-T₂ accuracy was high and insensitive to rejected navigators (< 5% variation for T₁ and T₂). In vivo PARMANav T₁ and T₂ values were higher than routine values but less variable, both per subject and between subjects: cortex PARMANav T₁ = 1601 ± 48 ms/T₂ = 90.8 ± 5.0 ms vs routine T₁ = 1307 ± 108 ms/T₂ = 73.3 ± 8.0 ms, medulla PARMANav T₁ = 2044 ± 82 ms/T₂ = 90.3 ± 5.4 ms and routine T₁ = 1560 ± 122 ms/T₂ = 67.6 ± 5.8 ms. No T₁ or T₂ trend was observed for the different NAWW. High-quality maps were obtained in the patients.

Conclusion: With accuracy confirmed in the phantom study and precision demonstrated in volunteers, PARMANav allows for precise and accurate renal joint T₁-T₂ mapping during free-breathing while minimizing through-plane motion.

Deuterium metabolic imaging (DMI) is an emerging magnetic resonance technique that enables non-invasive investigation of in vivo metabolism without the use of ionizing radiation. By administering various deuterium-labeled substrates, different metabolic pathways and fluxes can be probed. To date, most DMI studies have focused on cerebral metabolism; however, its application is rapidly expanding to include metabolic processes in other body organs and tissues, as well as non-brain tumors. This review summarizes the current state of in vivo DMI research beyond the brain, covering studies of the liver, non-brain tumors, and other organs, such as pancreas, kidney, and heart. With ongoing methodological developments and increasing emphasis on clinical translation, DMI holds considerable promise as a versatile tool for studying human metabolism and for future clinical implementation.

Objective: To evaluate an accelerated free-breathing three-dimensional stack-of-radial MRI technique of liver proton density fat fraction (PDFF) and $\text{R}_2^{\ast }$ quantification on retrospectively undersampled pediatric patient data.

Methods: A newly developed compressed sensing reconstruction with multidimensional regularization was evaluated on undersampled data in 21 clinical pediatric subjects at 1.5 T with respect to the reference-standard self-gating reconstruction on oversampled data (700 radial views per slice). PDFF and $\text{R}_2^{\ast }$ maps were calculated using the proposed method with fully sampled and undersampled radial views (352 and 176) and compared to reference-standard results using Bland-Altman analysis (reported as [mean difference; lower, upper limits of agreement]). Practical equivalence was evaluated using Bayesian posterior analysis with region of practical equivalence (ROPE) at ± 3% for PDFF and ± 10 s^-1 for $\text{R}_2^{\ast }$ .

Results: Echo images, PDFF and $\text{R}_2^{\ast }$ maps reconstructed using the proposed method had reduced image artifacts. Bland-Altman plots showed that the proposed method using 176 views had lower biases, [-0.7;-2.4,1.0]% for PDFF and [-5.3;-25.5,14.9] s^-1 for $\text{R}_2^{\ast }$ , compared to the reference method using 352 views: [0.8; -0.6, 2.1]% for PDFF and [-9.2;-51.9,33.4] s^-1 for $\text{R}_2^{\ast }$ . Bayesian posterior analysis revealed that our proposed method using 176 views was practically equivalent to the reference method using 700 views with 100% within the PDFF ROPE and with 98.25% within the $\text{R}_2^{\ast }$ ROPE. An acceleration factor of 4 and an approximate acquisition time saving of 67% shorter could be achieved.

Conclusion: The proposed method may allow accelerated free-breathing liver PDFF and $\text{R}_2^{\ast }$ mapping in pediatric subjects in approximately 1 min.