To evaluate how data acquisition and post-processing choices influence ultrashort echo time (UTE)–based characterization of the short-T2* compositional structure of the Achilles tendon. Ten healthy volunteers (33.3 ± 6.9 years) underwent Achilles tendon MRI using a multi-echo, time-interleaved UTE sequence. T2* data were analyzed with and without fat suppression using two models: (i) a bi-component decay with two water pools and (ii) a tri-component decay including two water components and one fat component. The effect of magnitude versus complex fitting on compositional estimates was also assessed. Three Achilles tendon regions were analyzed: insertion (INS), mid-portion (MID), and muscle–tendon junction (MTJ). Estimated short vs. long T2* components differed significantly depending on the use of fat suppression (p < 0.001 for all tendon regions) and fitting method (magnitude vs. complex) (pINS = 0.002, pMID < 0.001, pMTJ < 0.001). T2* time constants were comparatively stable, particularly in the MID part (p > 0.05). The fitting method affected short T2* component in the MTJ (pMTJ = 0.032) and long T2* component in the INS and MTJ (pINS = 0.025, pMTJ < 0.017). Non-fat-suppressed vs. fat-suppressed (FS) strategy and fitting approach substantially affect bi-component T2* quantification of the Achilles tendon, limiting comparability of UTE-based metrics. Tri-component modeling provides a more realistic description of tendon signal and may improve sensitivity to pathological changes. Further studies in patients are needed.
VERDICT MRI leverages the diffusion time-dependence of the diffusion MRI signal to estimate microstructural parameters in tumors. However, estimated parameters may be biased due to differences in T2 relaxation times between microstructural compartments, which are often neglected in conventional modelling. This study investigates how choice of echo time affects the estimated parameters in brain tumors and evaluates the use of relaxation-VERDICT, which incorporates T2 relaxation into the model to potentially reduce estimation biases and improve tumor microstructure estimation. Ten adult patients with intracranial tumors (mean age 57 years, range 32–78 years, 5 males) underwent diffusion MRI using a dual-echo acquisition with varying echo times, b-values, and diffusion times. VERDICT model fitting was performed using datasets with either a fixed echo time or a variable minimum echo time, and with models that either included or excluded compartment-specific T2 relaxation. Our results show that VERDICT parameter estimates vary significantly with the inclusion of T2 relaxation in the model, e.g. leading to a substantial increase of intracellular volume fraction. Estimated T2 relaxation was significantly shorter for the intracellular space (median 65 ms, IQR 60–70 ms) than for the extracellular space (median 246 ms, IQR 218–252 ms). Results indicate that accounting for T2-related echo time effects is important for proper biophysical interpretation of estimated VERDICT parameters and may thereby improve diagnostic reliability for brain tumors.
OBJECTIVE:The accurate segmentation of bone and cartilage is important for knee joint assessment. However, the complex shapes of both tissues make segmentation challenging. In this study, a conditional generative adversarial network incorporating dual attention mechanisms is proposed to enhance the segmentation accuracy of bone and cartilage. MATERIALS AND METHODS:The generator (a U-Net) integrates coupling attention mechanisms: squeeze-and-excitation in both the downsampling and upsampling paths, and the attention gate in skip connections, to focus on small tissue regions effectively. The discriminator (a convolutional neural network) fuses feature information from different layers of the generator. The proposed method was trained and tested using images from the Osteoarthritis Initiative (OAI) database. The training set and testing set consisted of 70 and 30 subjects, respectively. RESULTS:The experimental results demonstrated that the proposed method achieved satisfactory segmentation performance for four types of tissues: the femur, tibia, femoral cartilage, and tibial cartilage. The Dice similarity coefficient (DSC) values of these tissues were 97.4%, 97.2%, 91.2%, and 88.9%, respectively. Compared with using only GAN, the DSC of femoral cartilage and tibial cartilage increased by 6.9 and 1.7%, respectively. DISCUSSION:By integrating squeeze-and-excitation and attention gate, the network achieved a notable improvement in segmentation accuracy for knee cartilage.
To evaluate intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) for preoperative diagnosis of perineural invasion (PNI) in rectal cancer (RC). A total of 148 patients with pathology-confirmed RC (PNI+, n = 72; PNI-, n = 76) were enrolled. Parameters from mono-exponential (ADC), bi-exponential (D, D*, f), and stretched-exponential (DDC, α) IVIM models were analyzed. Univariate and multivariate logistic regression analyses were used to construct diagnostic models. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis. The DeLong test was used to compare the AUC of the models. Internal validation was employed to assess model performance. Net reclassification improvement (NRI) and integrated discrimination improvement (IDI), along with calibration metrics and decision curve analysis, were used to further evaluate model performance. P-value < 0.05 was considered statistically significant. ADC, D, f, and DDC differed significantly between groups. Multivariate analysis identified ADC and D as independent PNI predictors. The D value yielded the highest AUC (0.84), while ADC showed the highest sensitivity (81.94
Although undersampling combined with deep learning (DL)-based reconstruction shortens MRI acquisition, it increases the chance of inaccuracies, highlighting the need for quantifiable uncertainty measures. Two inference-time perturbation strategies, echo-train dropout (ET-Drop) and Gaussian noise Monte Carlo sampling (GN-MC), were compared in terms of the correlation between their variance-based uncertainty maps and absolute reconstruction error in DL-accelerated T2w prostate MRI. This retrospective multi-center study used a publicly available dataset with 312 k-spaces from NYU for training and a dataset with 120 k-spaces from University Medical Center Groningen for external validation. Fully sampled 3 T data were retrospectively undersampled to acceleration factors R = 3 and R = 6 and reconstructed by a vSHARP model. Per slice, five GN-MC perturbations were reconstructed by adding complex noise at 2.5σ, and five ET-Drop perturbations, created by omitting non-central echo trains. Voxel-wise aleatoric uncertainty was defined as the variance (σ2) across these reconstructions and correlated with absolute reconstruction error over whole slices and within the prostate. Both uncertainties yielded moderate slice-level correlations with absolute error. At R = 3, ET-Drop slightly outperformed GN-MC (median ρ = 0.39 vs 0.35; p < 0.001). At R = 6, the ranking reversed (0.44 vs 0.40; p < 0.001). Correlations within the prostate fell to 0.10–0.15. ET-Drop variance maps were dominated by coil sensitivities. Both perturbation strategies yield variance-based uncertainty maps that correlate moderately with voxel-wise error. More importantly, they consistently highlighted acquisition-related fragility, supporting the role of uncertainty mapping as a useful quality-control tool in prostate MRI.
To develop a simple fat-suppression strategy for more accurate quantitative ultrashort echo time magnetization transfer (UTE-MT) imaging of knee joint tissues. A narrow-bandwidth RF pulse centered on the fat peak was utilized for fat-selective imaging. These fat-selective images were subsequently subtracted from the MT-weighted 3D-UTE images to mitigate fat contamination in knee joint tissue imaging. This method was evaluated in an ex vivo human knee specimen and in three healthy volunteers at 3 T. Voxel-wise and ROI-based quantitative MT modeling were both performed to estimate macromolecular fraction (MMF), and results with and without fat suppression were compared. Fat-selective images revealed off-resonance artifacts from surrounding fatty tissues that extended into adjacent joint structures. Subtraction of the fat-selective images effectively suppressed these artifacts and improved visualization of cartilage, meniscus, and tendons. In ex vivo data, fat suppression stabilized MMF estimates by correcting voxels with abnormally elevated values (> 50
Fetal brain MRI is extremely challenging due to motion and relies almost exclusively on 2D single-shot imaging with multi-stack acquisition. 3D slice-to-volume reconstruction (SVR) has been used to improve through-plane resolution and conspicuity and compensate for the fetal motion. Lower field strengths such as 0.55 T are emerging as an important tool for fetal evaluation due to improved safety and patient comfort, reduced cost, and potential for better access. This study aims to determine the impact of 0.55 T fetal brain MRI acquisition parameters on fetal brain SVR reconstruction performance and determine optimal settings. We recruited nine healthy pregnant women (ten fetuses). Fetal brain imaging was performed at 0.55 T using T2-weighted half-Fourier acquisition single-shot turbo spin echo (HASTE) sequence in three orthogonal orientations. We acquired scans for four echo time (TE) values (98,140,181, and 272 ms) with 12 stacks (4 in each major orientation, fetal-axial, fetal-sagittal, and fetal-coronal). SVR was performed for 2 to 12 stacks for four the TE values; the results were evaluated qualitatively and quantitatively by a pediatric radiologist. With increasing TE, WM-GM contrast improves, and WM and GM SNR efficiency decreases, as expected. As the number of stacks increases, the SNR, structural similarity, and normalized error of SVR reconstruction monotonically improve, as expected. A TE of 140 ms, coupled with six stacks, provided radiologist preferred contrast to noise, with the shortest possible scan time. SVR is applicable to fetal brain MRI at 0.55 T and benefits from a longer TE and larger number of stacks than is typically used at 1.5 T or 3 T.
Q-space trajectory imaging (QTI) enables detailed characterization of tissue microstructure. Achieving high spatial resolution is challenging due to low signal-to-noise ratio (SNR), particularly on clinical MRI systems with limited gradient capabilities and coil options. This study assessed the potential of denoising methods to improve the resolution of QTI in the brain on a radiotherapy-dedicated MRI scanner. Using a 3T scanner with a 33 mT/m gradient system, we evaluated four denoising approaches: three methods based on principal component analysis (PCA) and Air Recon DL. Diffusion MRI of phantom and in vivo brain was acquired at voxel sizes from 3 × 3 × 3 to 1.25 × 1.25 × 1.25 mm3 using both diagnostic and radiotherapy coil setups. Precision and bias were analyzed, leading to in vivo brain QTI tested at resolutions 2 × 2 × 3 (radiotherapy coil) and 2 × 2 × 2 mm3 (diagnostic coil). Denoising complex images was required to mitigate noise floor bias and increase resolution. The denoising methods varied in performance in terms of signal variance at high b-values. One PCA method enabled a decreased voxel size of 2 × 2 × 3 (radiotherapy coil), improving the parameter contrast-to-noise ratio, particularly for fractional anisotropy (+ 30
Single-voxel proton magnetic resonance spectroscopy (1H-MRS) is a non-invasive in vivo imaging technique used to quantify the concentration of human brain metabolites. Frequency-selective 1H-MRS techniques reduce spectral complexity and simplify spectral modeling. We introduce a single-shot frequency-selective sequence known as Delays-Alternating-Nutation-Tailored-Excitation-Point-RESolved-Spectroscopy (DANTE-PRESS) and test its precision when measuring glutamate and NAA at 7 Tesla in phantoms and human brain in vivo. DANTE-PRESS was programmed within the software environment of the Siemens Magnetom 7 Tesla MR scanner at the Centre for Functional Metabolic Mapping in London, Ontario. Two scans were obtained on phantoms and in 4 healthy volunteers (20 × 20x20mm3 voxel at the dorsal anterior cingulate cortex) as an in vivo proof-of-concept to refocus glutamate or NAA. DANTE-PRESS preserves the signal of the metabolite of interest while suppressing unwanted signals via a narrow-band frequency-selective refocusing pulse. DANTE-PRESS produces metabolite spectral signatures with J-evolution equivalent to that seen in a PRESS sequence with less than half the echo time. The inter-individual coefficients of variance were low and Cramer-Rao Lower Bounds were less than 5.1
Pulmonary MRI in neonates can be performed with quality comparable to radial 3D ultrashort echo time (UTE) MRI in significantly less time using a FLORET trajectory. Eighteen NICU patients with severe bronchopulmonary dysplasia (BPD), age 40.9 ± 3.0 weeks at time of imaging, underwent MRI using Radial and FLORET UTE at 1.5 T. Pulmonary signal-to-noise ratio (SNR), lung density, and radiologist scoring of motion artifacts and image quality were compared across sequence types. FLORET UTE reduced scan time by 75
OBJECTIVES:This study aimed to develop a super lightweight deep learning model for brain age estimation using structural MRI, enabling accurate age estimation with minimal computational cost for deployment in resource-constrained clinical settings. METHODS:A super lightweight brain age estimation network, termed superLPNet, was proposed. Lightweight convolutional structures inspired by MobileNet were adopted to reduce model parameters and computational burden. Spatial and channel attention mechanisms were further integrated to enhance feature representation without substantially increasing model complexity. The proposed model was evaluated on a combined dataset of 3550 T1-weighted MRI scans and further validated on an independent Alzheimer's disease (AD) cohort. RESULTS:The superLPNet achieved the lowest mean absolute error compared with state-of-the-art models, demonstrating superior brain age estimation accuracy. The number of parameters was reduced by 56.70%-98.75% relative to competing approaches, highlighting its super lightweight design. From a clinical perspective, patients with AD exhibited a significantly larger brain age gap than healthy controls. CONCLUSIONS:The proposed model enables accurate and efficient brain age estimation using T1-weighted MRI with substantially reduced complexity, supporting its potential for real-world clinical application.
Diffusion-weighted imaging (DWI) has been adopted to study placentas of women infected by SARS-CoV-2, highlighting the microstructural deterioration of the tissues using the Intravoxel Incoherent Motion model (IVIM). However, unlike the results obtained on placental histology, no perfusion impairment was observed. We used the Two-perfusion IVIM model to investigate the placenta's perfusion compartments. DWIs of n = 12 patients affected by SARS-CoV-2 and n = 20 gestational age-matched pre-pandemic healthy subjects were acquired at 1.5 T scanner with 10 b values. The maternal and fetal placentae were investigated. Differences in the two groups were evaluated using Welch’s t test. The placental fetal side showed higher values of D and lower values of f2 (relative to the trophoblast compartment) in SARS-CoV-2 patients. D2* was higher on the maternal side of SARS-CoV-2 patients. Higher values of D reflect damage to the tissue, while the lower f2 parameter suggested an impairment of the exchange between mother and fetus. The higher values of D2* in the maternal side of SARS-CoV-2 placentas may be due to a decreased capillaries’ size caused by the infiltration of lymphocytes in the decidua basalis. These results further confirm the potential of the two-perfusion IVIM model in detecting placental dysfunction.
To provide a thorough comparison of the SNR between sodium MRI k-space sampling schemes in the brain within clinically feasible time constraints (∼10 min) at 3 T. Density-adapted radial (DA-3DPR), constant-amplitude radial, Cartesian, FLORET, rotated spiral, and 3D cones trajectories were designed with parameters optimized for brain tissue SNR. The sequences were acquired in both a phantom and 13 healthy participants (age = 28.7 ± 3.4, M:F = 7:6). SNR was measured and corrected for point-spread function (PSF) volume and scan duration for a less-biased assessment. CSF-to-brain-tissue contrast and CNR were also measured. The data were linearly modeled, and ANOVA was used to determine if the sampling scheme contributed to the variance with the obtained metrics. The sampling schemes contributed significantly to the variance (p < 0.001) for all metrics. The DA-3DPR sampling scheme provided the highest SNR in both the phantom and the participants. The Cartesian sampling scheme had the highest absolute contrast, but the largest CNR was shared between the DA-3DPR, 3D cones, and FLORET sampling schemes. When considering the PSF and the requirement for a clinically feasible scan time, a 15 ms read-out DA-3DPR trajectory provides the highest SNR at 3 T, without losing any desired contrast.
Gliomas are heterogeneous brain tumors with variable biology and treatment response. Accurate, non-invasive assessment of tumor aggressiveness is essential for prognosis and treatment planning. Conventional machine learning (ML) approaches typically frame glioma grading as a discrete classification task, which may overlook substantial intra-grade heterogeneity. This pilot study explores a regression-based framework to derive a continuous imaging-derived severity score, providing a relative assessment of tumor aggressiveness anchored to, but not redefining, established WHO grades. 36 glioma patients (low-grade glioma; LGG: 58.33
To introduce and evaluate a novel multi-contrast dual-resolution 3D-UTE sequence (multi-UTE) for cerebral myelin fraction (MF) estimation in vivo and benchmark it against electron microscopy-based myelin volume fraction. Direct detection of myelin in tissue is challenging due to the ultrashort T_2^* of the myelin bilayer. In this work, a high-resolution UTE pulse sequence with 1–1 binomial water excitation pulses was combined with interleaved multi-echo UTE acquisitions at lower resolution, enabling simultaneous acquisition of quantitative, water-excited, high-resolution T1-weighted structural images. The study included two ex vivo sheep brains, five healthy volunteers, and three patients with multiple sclerosis (MS) were measured. MF maps were calculated using multi-UTE and compared to histology. Magnetization transfer ratio using routine sequences was also calculated as a reference. Multi-UTE facilitates the simultaneous acquisition of quantitative and structural images in 20 min. Myelin fraction values tested in five healthy volunteers agreed well with the literature. The MF estimated using multi-UTE showed a high correlation with histology (R2 = 0.86) across 120 samples from different brain regions. Multi-UTE showed the strongest correlations with electron microscopy (EM) histology, and the MF maps overlapped with the lesions in MS patients. Multi-UTE may provide complementary insights for myelin quantification in various inflammatory and neurodegenerative diseases.
Isocitrate dehydrogenase (IDH) mutations are key prognostic factors in gliomas. IDH-wildtype (IDH-wt) glioblastomas are high-grade tumors, which often exhibit blood–brain barrier (BBB) breakdown. Blood–brain barrier arterial spin labeling (BBB-ASL) is a novel MRI technique that provides information on cerebral blood flow (CBF) and water exchange time (Tex) across the BBB. The aim of this study was to evaluate the feasibility of using BBB-ASL to measure CBF and Tex for distinguishing IDH mutational subgroups in gliomas, while accounting for region-specific T2 relaxation time variations. Twenty-five histopathologically confirmed gliomas (15 IDH-wt and 10 IDH-mutant (IDH-mut); mean age 53.6 ± 14 years; F/M = 11/14) were scanned. Hadamard-4 and Hadamard-8 pseudo-continuous arterial spin labeling (pCASL) MRI data were preprocessed in ExploreASL using the BBB-ASL model. CBF and Tex maps were generated using fixed T2 values (tissue: 85 ms; blood: 165 ms), and recomputed with a tissue T2 of 116.2 ms (CBF_116ms and Tex_116ms) and patient specific regional median T2 (CBF_corr and Tex_corr). Relative CBF and Tex maps (rCBF_116ms, rTex_116ms; rCBF_corr, rTex_corr) were calculated by normalizing to median normal-appearing gray matter (NAGM) values. Group comparisons of rCBF and rTex values between IDH mutational subgroups were performed using Mann–Whitney U tests with Holm-Bonferroni correction for multiple comparisons. rTex_116ms maps showed significantly lower energy in IDH-wt tumors ( P=0.010 ) and a higher 95^th percentile ( P=0.049 ) in IDH-mut tumors. In addition, IDH-wt gliomas demonstrated a trend toward higher 5^th percentile rCBF_116ms and rCBF_corr ( P=0.033 , P=0.014 , respectively). Multi-TE BBB-ASL provides noninvasive insight into perfusion and water exchange characteristics in gliomas, with T2 relaxation time assumptions strongly influencing CBF and Tex estimates.
OBJECTIVE:To test the hypothesis that idiopathic normal-pressure hydrocephalus (iNPH) is also associated with reproducible whole-brain structural connectivity reductions detectable by diffusion MRI connectomics, beyond the established clinical triad of gait disturbance, cognitive impairment, and urinary incontinence. MATERIALS AND METHODS:Twenty patients with iNPH (71-84 years; Evans index 32.1-43.5) diagnosed according to the Japanese guidelines and 20 age-matched healthy controls (70-88 years) underwent 3.0-T MRI. Distortion-corrected diffusion-weighted imaging (b = 2000s/mm2, 32 directions) was analyzed using constrained spherical deconvolution and anatomically constrained probabilistic tractography. Connectivity matrices were constructed for 84 regions (Desikan-Killiany atlas), yielding 3486 unique edges. Group differences were tested edge wise using the Mann-Whitney U test with Bonferroni-adjusted p-values; statistical significance was defined as adjusted p < 0.01, and effect sizes were quantified using the rank-biserial correlation (large effect: r > 0.5). RESULTS:iNPH showed significantly reduced connectivity predominantly in interhemispheric edges (adjusted p < 0.01; r = 0.89-0.98). The paracentral lobule exhibited the greatest number of decreased connections (11 edges), followed by the posterior cingulate gyrus (9 edges) and the superior frontal gyrus (8 edges). CONCLUSIONS:iNPH is associated with robust interhemispheric disconnection and prominent involvement of motor- and midline-related networks, supporting connectome-derived markers for further validation.
In quantitative dynamic contrast-enhanced MRI (DCE-MRI), a fundamental trade-off exists between imaging speed, spatial resolution, and signal-to-noise ratio (SNR), driven by the amount of data acquired per dynamic frame. This work proposes a model-based reconstruction (MBR) framework for directly estimating pharmacokinetic parameters from raw k-space data, eliminating the need for intermediate image reconstruction and potentially mitigating this trade-off. The extended Tofts model for pharmacokinetic modeling was integrated into an MBR framework—PyQMRI (code is shared). To validate the approach in a controlled setting, a simulated digital phantom of the abdominal region was used. Pharmacokinetic parameters were generated, and corresponding k-space data were calculated based on these values. Additionally, the feasibility of MBR was evaluated in vivo using liver DCE-MRI data from a healthy volunteer. The performance of MBR was compared to a conventional image-based fitting approach using a non-linear least-squares (NLLS) algorithm. In simulations, MBR showed superior performance, producing more precise pharmacokinetic maps with accuracy comparable to or exceeding that of traditional image-based fitting. Notably, fine anatomical structures, such as blood vessels, were more clearly defined with MBR. This improvement was consistent across different temporal resolutions (1.5–16.5 s/frame). MBR did not show any sign of image degradation for shorter frame rates, and, in fact, performed best with the shortest tested frame rate (1.5 s/frame), highlighting the robustness of MBR image quality to higher frame rates. In vivo, while the improvements offered by MBR were consistent with simulation results, they were less pronounced. Several aspects could have contributed to this discrepancy, including a difference between the simple extended Tofts model and the complex true in vivo pharmacokinetics, and data degradation due to motion combined with a complex landscape of the DCE loss. The proposed MBR framework offers a promising alternative to traditional DCE-MRI workflows by avoiding intermediate image reconstruction and relaxing the spatio-temporal trade-off. This approach may enable more accurate and robust estimation of pharmacokinetic parameters, particularly in scenarios where imaging constraints are severe.
The purpose of this study was to implement a nonmagnetic micropositioner in the MRI environment to validate displacement estimates of magnetic resonance acoustic radiation force imaging (MR-ARFI). The micropositioner consisted of a stage driven by a piezoelectric stepper motor in closed-loop operation with an optical quadrature encoder. A 100-gram agar gel phantom was prepared, and three MR-ARFI pulse sequences were used to generate displacement maps. MR-ARFI measured displacements were compared to ground truth data from the optical encoder. The micropositioner demonstrated consistent performance with positioning times of 1.6 ± 0.4 ms for extension and 2.2 ± 0.2 ms for return to baseline position. The micropositioner decreased the signal-to-noise ratio of magnitude images due to increased electronic noise. Linear regression analysis showed that displacement measurements were highly linear with R^2 ≥ 0.98 but exhibited scaling biases that may have been due to the experimental setup. The proposed instrument can potentially improve the accuracy and precision of MR-ARFI-based applications, including focused ultrasound dosimetry and mechanical biomarker imaging.
OBJECTIVE:To investigate the potential of the unidirectional cellular water efflux rate constant (kio) from DCE-MRI data in prostate MR imaging. MATERIALS AND METHODS:High-temporal-resolution prostate DCE-MRI data were modeled using both the fast-exchange-limit (FXL) Tofts' model as well as the water-exchange-sensitized shutter-speed model (SSM). In the SSM, kio was included as an additional fitting parameter. Lesion and normal-appearing (NA) prostate tissue region-of-interest (ROI) data were analyzed and categorized into FXL or non-FXL conditions based on results from the two models. A global upper limit of kio detectable by prostate DCE-MRI with the SSM was presented. RESULTS:While many lesion voxels exhibited sensitivity to kio with the SSM, a substantial portion remained in the FXL condition despite greater contrast agent extravasation than in NA tissue. The fraction of FXL voxels was higher in lesions than in NA tissue. Applying a global detectable kio upper limit increased the difference between lesion and NA ROIs, improving lesion characterization. DISCUSSION:SSM-derived FXL and non-FXL contrasts may serve as novel imaging biomarkers for prostate cancer surveillance. Advances in MRI technology and more potent contrast agents are expected to enhance the accuracy of kio quantification, potentially enabling its integration into clinical mpMRI.
