OBJECTIVE:To develop a deep learning algorithm for semiquantification of spinal inflammation in patients with axial spondyloarthritis (SpA). METHODS:The study included 330 participants with axial SpA. All patients underwent whole spine MRI with short τ inversion recovery (STIR) sequence by 3T MR unit. Three independent readers identified regions of interest to locate bone marrow oedema (BMO) and performed Spondyloarthritis Research Consortium of Canada (SPARCC) scoring. Two deep learning models based on attention Unet were developed. The BMO model differentiated image with or without spinal inflammation. The vertebral body (VB)-intervertebral disc (IVD) model identified discovertebral units for localisation. The intraclass correlation coefficient (ICC) and Pearson coefficient were used to evaluate agreement and correlation between scorings by human readers and deep learning-based pipeline. Performance of the models was evaluated using sensitivity, specificity, accuracy and Dice coefficient. RESULTS:The ICC and the Pearson coefficient of SPARCC scores between human readers and the deep learning-based scoring pipeline were 0.80 and 0.82, respectively. The sensitivity and specificity of spinal inflammation identification were 0.90 and 0.84, respectively. The Dice coefficients were 0.81 (VB) and 0.80 (IVD) in images with spinal inflammation. CONCLUSION:The high consistency of the scoring pipeline with human readers suggested that the deep learning-based algorithm has the potential to provide semiquantitative assessment of spinal inflammation based on SPARCC in axial SpA.

Odor information is transmitted from the olfactory bulb to several primary olfactory cortical regions in parallel, including the anterior olfactory nucleus (AON) and piriform cortex (Pir). However, the specific roles of the olfactory bulb and cortical outputs in wider interactions with other interconnected regions throughout the brain remain unclear due to the lack of suitable in vivo techniques. Furthermore, emerging associations between olfactory-related dysfunctions and neurological disorders underscore the need for examining olfactory networks at the systems level. Using optogenetics, fMRI, and computational modeling, we interrogated the spatiotemporal properties of brain-wide neural interactions in olfactory networks. We observed distinct downstream recruitment patterns. Specifically, stimulation of excitatory projection neurons in OB predominantly activates primary olfactory network regions, while stimulation of OB afferents in AON and Pir primarily orthodromically activates hippocampal/striatal and limbic networks, respectively. Temporally, repeated OB or AON stimulation diminishes neural activity propagation brain-wide in contrast to Pir stimulation. Dynamic causal modeling analysis reveals a robust inhibitory effect of AON outputs on striatal and limbic network regions. In addition, experiments in aged rat models show decreased brain-wide activation following OB stimulation, particularly in the primary olfactory and limbic networks. Modeling analysis identifies a dysfunctional AON to Pir connection, indicating the impairment of this primary olfactory cortical circuit that disrupts the downstream long-range propagation. Our study delineates the spatiotemporal properties of olfactory neural activity propagation in brain-wide networks for the first time and distinguishes the roles of primary olfactory cortical, AON and Pir, outputs in shaping neural interactions at the systems level.

Four-dimensional magnetic resonance fingerprinting (4DMRF) provides multi-parametric and motion-resolved tissue property quantification, promising to enhance the precision of liver cancer radiotherapy. However, its clinical translation is hindered by prolonged reconstruction time. Deep learning acceleration is fundamentally constrained by the lack of ground-truth 4D data. To this end, we propose SS-4DMRF, the first self-supervised reconstruction framework for 4DMRF, to reconstruct motion-resolved tissue maps without using supervised image labels. SS-4DMRF features a core temporal low-rank-constrained registration (TelReg) network for precise motion modeling. It leverages the intrinsic low-rank compressibility of respiratory motion and is directly self-supervised by the original highly undersampled k-space data and the derived subspace images. Motion-resolved tissue maps are reconstructed using a motion-informed compensation approach via a physics-informed pattern matching (PiPM) network. PiPM network incorporates novel multi-scale Swin Transformers with Bloch-equation-guided subspace denoising to achieve high-fidelity tissue quantification. SS-4DMRF was validated on digital phantom (n=30) and in vivo liver cancer patient (n=33) datasets. Compared to state-of-the-art 4DMRF methods, SS-4DMRF demonstrated superior tissue quantification and motion measurement accuracy. It achieved significantly reduced NRMSE in 4D tissue property quantification and improved inter-phase structural repeatability in 4D motion characterization (Paired Student's t-tests, p<0.001). The measured tumor motion trajectory presented strong Peason correlation with motion reference (r=0.939±0.057). Crucially, SS-4DMRF achieves this dual improvement in accuracy with a 10-fold acceleration in reconstruction time compared with conventional 4DMRF methods. By enabling rapid, precise, and motion-resolved quantitative imaging, SS-4DMRF advances the precision of liver cancer radiotherapy and establishes a clinically feasible platform for abdominal quantitative MRI in oncology.

Background:Point-of-care ultrasound training is being increasingly integrated into undergraduate medical education, leading to a substantial demand for trained faculty to provide instruction and feedback. Objective:This study aimed to develop an adjunct tool, a deep learning-based feedback model, to facilitate student learning. Methods:Renal ultrasound images (N=2807) were used to train a cascaded deep learning-based feedback model that classified images into three categories: optimal, suboptimal, and incorrect. Suboptimal images were further subcategorized as images with artifact, incorrect gain, and/or incorrect positioning. The model was deployed among year 5 medical students receiving bedside ultrasound training, who were invited to upload renal ultrasound images to an online platform for automated image quality grading and feedback. A mixed methods analysis was used to evaluate students' learning experience. Focus group interviews were organized to qualitatively analyze the successes and challenges of implementation. Quantitative analysis was based on responses to a 5-point Likert scale questionnaire and performance on the objective structured clinical examination (OSCE). Objective structured clinical examination scores were compared with mean OSCE scores from the 2 years preceding implementation of the deep learning-based feedback model. Results:Focus group interviews identified that the deep learning-based feedback model encouraged self-regulated learning but also recognized that discordant curricular design and hardware limitations impeded its use. The 11-item online questionnaire had a response rate of 42.4% (98/231 students). Among respondents, 32% (31/98) to 48% (47/98) found the model helpful in assisting ultrasound training (Likert score of 4-5 for items 1-3), while 49% (48/98) to 76% (74/98) were satisfied with its usability and their interaction with the model (Likert score of 4-5 for items 4-11). The mean OSCE score was 9.73 (SD 0.76) out of 10, compared with mean scores of 9.35 (SD 1.03; P=.06) and 9.45 (SD 0.97; P=.15) out of 10 in the 2 individual years preceding implementation of the model. Conclusions:A cascaded deep learning-based feedback model was developed and deployed among year 5 medical students receiving bedside ultrasound training, with positive learner responses and enhanced self-regulated learning. The innovation was associated with increased student engagement and improved ultrasound skill acquisition among novice learners.

Objective.This study aims to develop a motion-robust magnetic resonance fingerprinting (MR-MRF) technique for liver cancer imaging to eliminate the need for breath-hold scanning.Approach.To mitigate respiratory motion artifacts in free-breathing abdominal MRF, the MR-MRF technique comprising two core components. First, respiratory motion is modeled by applying an isotropic total variation (TV)-regularized registration algorithm between a target end-of-exhalation (EOE) phase and three motion phases. Second, motion-resolved tissue property maps are reconstructed using a low-rank TV optimization framework, which incorporates the estimated inter-phase motion to align all acquired MRF dynamics to the EOE phase. MR-MRF is evaluated by 22 patients (mean age, 62 years ± 10 [SD]; 15 males and 7 females) with hepatocellular carcinoma. Radiologist's blinded assessment and organ boundary sharpness measurements are performed to evaluate the image quality of MR-MRF-derived tissue maps. The test-retest tissue quantification repeatability is assessed by two consecutive MRF scans with distinct breathing patterns. Paired Student'st-test is used for statistical significance analysis with ap-value threshold of 0.05.Main results.MR-MRF achieved successful reconstruction of motion-resolved tissue maps at EOE phase, with blinded radiologist assessment yielding an average score of 3 (moderate quality-sufficient for diagnosis) for overall image impression. The FWHM of organ boundaries in MR-MRF-derived tissue maps is 3.1 mm ± 1.7 mm, significantly lower than motion-blurred tissue maps (9.9 mm ± 3.4 mm,p-value < 0.0001). Test-retest analysis demonstrated good repeatability: liver coefficient of variation was 5.5% ± 7.1% (T1), 8.2% ± 4.4% (T2), and 5.0% ± 2.0% (PD), with excellent linear agreement (R2= 0.96, 0.80, and 0.85 for T1, T2, and PD, respectively).Significance.This study establishes the technical foundation of MR-MRF to achieve repeatable and quantitative liver T1/T2/PD mapping under free-breathing conditions at 3 T. The results validate the feasibility of addressing respiratory motion in abdominal multi-parametric quantitative MRI.

PURPOSE:To develop a high-quality 4-dimensional magnetic resonance fingerprinting (HQ-4DMRF) framework with temporal low-rank-constrained motion compensation for precise tumor motion management in liver radiation therapy. METHODS AND MATERIALS:HQ-4DMRF integrated 4 key innovations: (1) an automated internal respiratory navigator to track organ motion without external sensors; (2) a results-driven phase-sorting algorithm to dynamically redistribute magnetic resonance fingerprinting (MRF) dynamics across respiratory phases; (3) a novel temporal low-rank-constrained 4-dimensional (4D) registration algorithm to simultaneously compute all interphase deformation vector fields by leveraging low-rank respiratory motion properties and enforcing spatiotemporal regularization; and (4) an iterative motion-compensated optimization algorithm to reconstruct motion-resolved 4D tissue maps. HQ-4DMRF was validated in 24 patients with hepatocellular carcinoma. All patients underwent a free-breathing abdominal MRF scan using a multislice 2-dimensional fast acquisition with steady-state precession sequence. The motion measurement accuracy of HQ-4DMRF was assessed through interphase structural repeatability. Interphase structural repeatability quantified the structural consistency in tissue maps across motion phases using the structural similarity index, local cross-correlation, and textural feature intraclass correlation coefficient for tumors. RESULTS:The HQ-4DMRF demonstrated superior precision in motion measurement versus conventional 4DMRF techniques (P < .001), with interphase structural repeatability-structural similarity index/-local cross-correlation/-textural feature intraclass correlation coefficient of 0.82 ± 0.06/0.36 ± 0.07/0.75 ± 0.20 for T1, 0.89 ± 0.05/0.29 ± 0.06/0.84 ± 0.24 for T2, and 0.80 ± 0.06/0.38 ± 0.06/0.91 ± 0.12 for proton density maps. Compared with using conventional pair-wised registration methods, the temporal low-rank-constrained 4D registration improved motion measurement accuracy by an average of 8.5% to 12.5% (structural similarity index), 9.1% to 36.2% (local cross-correlation), and 8.2% to 17.1% (textural feature intraclass correlation coefficient). The respiratory curve derived from automated internal respiratory navigator showed strong agreement with manual measurements (Pearson correlation coefficient = 0.90 ± 0.12) and demonstrated consistent performance across different anatomic regions (Pearson correlation coefficient = 0.83 ± 0.13). The result-driven phase sorting enhanced the 4DMRF performance by 7.2%. CONCLUSIONS:The HQ-4DMRF framework presents a comprehensive solution to critical challenges in 4DMRF. Clinical validation in patients with hepatocellular carcinoma demonstrates significant improvements in liver tumor motion characterization. These advances not only enhance the precision of radiation therapy planning through more accurate motion modeling but also establish HQ-4DMRF as a promising platform for 4D quantitative magnetic resonance imaging in oncologic applications.

BACKGROUND:Diagnosis and treatment of glioblastoma (GBM) rely on multiparametric MRI (mpMRI), but mpMRI is time-consuming and costly. Deep learning-based synthesis methods have been proposed to streamline acquisition; however, their generalizability is limited by variability in qualitative input contrasts across sites and scanners. PURPOSE:To overcome this limitation, we developed and evaluated a generalizable deep learning model that synthesizes mpMRI contrasts directly from quantitative magnetic resonance fingerprinting (MRF) maps in GBM patients. The proposed Quantitative Synthesis Network (QS-Net) employs a deeply supervised residual U-Net generator within an adversarial framework, combined with a two-stage training strategy to separate anatomical and pathological learning. METHODS:We collected MRF-derived T1 and T2 maps, along with conventional mpMRI sequences (T1w, T2w, T1-FLAIR, T2-FLAIR, and SWI), from 32 healthy volunteers and retrospectively from 18 GBM patient scans. The proposed QS-Net was initially trained on healthy volunteer data (20 scans for training, 12 for testing) to learn general anatomical features. Subsequently, it was fine-tuned using 9 GBM patient scans to adapt to pathological characteristics, with the remaining 9 patient scans reserved for independent testing. We compared the performance of QS-Net against three state of the art deep learning models: Res-Unet, conditional GAN, and Swin-Transformer, using both quantitative metrics (MAE, SSIM, and PSNR) and qualitative assessments. Additionally, we assessed the generalizability of the models by evaluating their external validation performance when trained with either conventional MRI or quantitative MRF inputs. RESULTS:QS-Net outperformed the comparison models in synthesizing T1w, T2w, SWI, and T2-FLAIR images for GBM patients, achieving the best results across all quantitative metrics: MAE (1.18 ± 0.52, 1.01 ± 0.36, 1.05 ± 0.37, 1.45 ± 0.76), SSIM (0.934 ± 0.037, 0.939 ± 0.039, 0.934 ± 0.034, 0.926 ± 0.053), and PSNR (29.69 ± 3.21, 29.35 ± 2.29, 29.64 ± 2.58, 27.56 ± 3.39), respectively. Qualitative analysis demonstrated that QS-Net generated synthetic images with superior resemblance to ground truth, accurately delineating tumor boundaries and preserving intra-tumoral texture. Furthermore, the generalizability test revealed that models trained on standardized quantitative MRF input maps consistently outperformed models trained on vendor-specific qualitative MRI inputs across all architectures and metrics (p < 0.005). CONCLUSIONS:We developed QS-Net, a deep learning model for high fidelity mpMRI synthesis from quantitative MRF maps, and demonstrated that this quantitative-input paradigm enables superior cross-vendor generalization over conventional qualitative MRI-based approaches.

BACKGROUND:Accurate evaluation of coronary artery constriction and myocardial ischemia is essential for diagnosing and managing coronary artery disease (CAD). Combining CT coronary angiography (CTCA) and stress cardiovascular magnetic resonance (CMR) imaging allows examination of both coronary artery narrowing and myocardial perfusion. PURPOSE:To develop a deep learning pipeline that integrates CTCA and CMR images, which could help improve accuracy in identifying affected vessels and their associated myocardial territories. METHODS:The proposed pipeline included two deep learning models: one for automatic reorientation of 3D CTCA and another for left ventricle (LV) wall registration between CTCA and CMR images. A 3D spatial co-registration model, the reorientation spatial transformer network (Reorientation STN), predicted reorientation parameters for input CTCA volumes using ResNet18 and STN. A 2D nonrigid spatial deformation network (Nonrigid SDN) was trained for LV wall registration. Cross-modal supervision was employed during training. Evaluation criteria included aspect ratio (AR), Dice similarity coefficient (DSC), and long-axis deviation angles. The process involved quantifying LV wall perfusion on registered CMR images and extracting coronary arteries from reoriented CTCA images to fuse these results. The pipeline was trained and validated on 447 pairs of CTCA and CMR images from 75 patients and tested on 18 subjects. RESULTS:The pipeline achieved an AR of 0.94 ± 0.03, long-axis deviation angles of 1.19 ± 0.83 (axial) and 1.54 ± 0.79 (coronal), a DSC of 0.66 ± 0.04 for LV wall reorientation, and a DSC of 0.92 ± 0.03 for LV wall registration between CTCA and CMR. CONCLUSIONS:This automated framework successfully fuses cardiac CTCA and CMR imaging, demonstrating its potential effectiveness.

BACKGROUND:Magnetic resonance fingerprinting (MRF) could provide joint T1, T2, and proton density mapping. Measuring diffusion encoding using the MRF framework is promising, given its capacity to generate self-aligned quantitative maps and contrast-weighted images from a single scan. It could avoid potential errors that arise from the registration of multiple MRI images and reduce the total scan time. However, the application of a strong diffusion gradient on the MRF sequence results in phase inconsistency between acquisitions, which could corrupt the reconstructed images. PURPOSE:To propose a distortion-free diffusion-weighted imaging module for MRF (DWI-MRF) method using a self-navigated subspace reconstruction on k-space data obtained from a dual-density spiral trajectory. METHODS:The proposed sequence consisted of two segments: inversion prepared steady-state free precession MRF for the first 800 time points and diffusion-weighted imaging (DWI) with two nominal b-values of 0 and 800 s/mm2 for the following 200 time points. The temporal basis was acquired from the densely sampled central k-space during reconstruction. The subspace reconstruction was applied to generate aliasing-free and high-resolution images at each time point. The cardiac gating was retrospectively performed on the high-resolution and dynamic DWI images. Our T1, T2, and apparent diffusion coefficient (ADC) results were compared to conventional methods on a phantom and two healthy volunteers. RESULTS:Our method's T1, T2, and ADC values agreed reasonably with the reference values, with a slope of 0.88, 0.94, and 1.04 for T1, T2, and ADC, and an R2 value of 0.97, 0.97, and 0.71, respectively. The T1, T2, and ADC maps from DWI-MRF exhibited pixel-by-pixel correspondence on phantom and in vivo (T1 and ADC: R2 = 0.75 on phantom and 0.84 in vivo; T2 and ADC: R2 = 0.79 and 0.83, respectively). Our method achieved high acquisition efficiency, requiring less than 20 s per slice. CONCLUSIONS:The proposed method was free of artifacts from cardiac pulsation and generated pixel-wise correspondent T1, T2, and ADC maps on both phantom and in vivo images.

Magnetic resonance imaging (MRI) signal acquisition relies heavily on radio frequency (RF) coils, which play a critical role in obtaining high-quality images. However, traditional RF coils are often rigid and bulky and require complex decoupling mechanisms, limiting their adaptability and effectiveness for anatomically curved or dynamic regions. In this study, we present a novel high-impedance nonlinear metasurface (HINM) coil based on a “building bricks” concept to overcome these limitations. The HINM coil features a lightweight, flexible, wireless, and compact design that enables direct attachment to various anatomical regions, providing localized signal enhancement without the need for intricate adjustments for decoupling. This innovative approach utilized a flexible coaxial cable with a shielded design, ensuring stable frequency characteristics under bending, stretching, and dynamic conditions. Furthermore, the HINM coil with passive detuning did not alter the RF transmit field distribution while improving the signal-to-noise ratio (SNR) up to 87% in the surface region, as demonstrated by the phantom studies. In the knee and hand imaging, the SNR in the joint and finger areas was also doubled when using the HINM arrays. Enhanced image quality was achieved, and more subtle blood vessels were revealed in the hand vascular imaging with the use of the HINM arrays in combination with the commercial RF coils. Additionally, the HINM arrays were applied for the knee and hand imaging at 0.5 T, which achieved up to 74% SNR improvement, demonstrating its effectiveness for both low-field and ultrahigh-field MRI. By offering a versatile and adaptable solution, the HINM coil demonstrates its potential to transform MRI coil designs, particularly for imaging anatomically complex and dynamic environments.

BACKGROUND:Multiple sclerosis (MS) is an autoimmune demyelinating disease that attacks myelin. MRI is an important imaging modality for diagnosis and monitoring in MS. However, the current standard MRI protocol for MS lacks sequences capable of detecting molecular changes. PURPOSE:To present a saturation-transfer-based MRI protocol, including chemical exchange saturation transfer (CEST) and magnetization transfer indirect spin labeling (MISL) sequences, for quantifying molecular changes and water exchange in the brain of MS patients. STUDY TYPE:Prospective. POPULATION:Fifty-two participants including 31 healthy controls (HC) (18 females and 13 males) and 21 MS patients (18 females and 3 males). FIELD STRENGTH/SEQUENCE:3D inversion-prepared gradient echo T1w, 3D fast spin echo T2w, 3D CUBE CEST and MISL at 3.0 T. ASSESSMENT:Multiple CEST contrasts between HC and MS groups were analyzed using double-step multi-pool Lorentzian fitting (DMPLF) and Lorentzian difference analysis (LDA) to evaluate and compare their diagnostic performance. MISL signals at -20 and -10 ppm were quantified by the normalized signal reduction in cerebrospinal fluid (CSF). T1w MRI was used to quantify brain volumes. STATISTICAL TESTS:Unpaired Student's t-test, receiver operating characteristic (ROC) curve, area under the curve (AUC), and binary logistic regression analysis. p < 0.05 was considered statistically significant. RESULTS:CEST detected decreased signals in the brain of MS patients using both DMPLF and LDA, with DMPLF demonstrating superior performance in differentiating MS from HC (AUC, 0.93; 95% CI: 0.86, 1.00). MS patients showed significantly lower whole brain MISL signals than HCs at both -20 ppm (0.04 ± 0.01 vs. 0.06 ± 0.02) and -10 ppm (0.06 ± 0.02 vs. 0.08 ± 0.02). MS patients showed a significant decrease (-6.57%) in brain tissue and an increase (+20.73%) in CSF volume ratios compared to HCs. DATA CONCLUSION:The saturation-transfer-based MRI framework can effectively evaluate molecular changes and CSF-tissue water exchange in the brains of MS patients. EVIDENCE LEVEL:2. TECHNICAL EFFICACY:Stage 3.

Background: Deep learning models based on the Spondyloarthritis Research Consortium of Canada (SPARCC) scoring system have been used in previous studies to assess sacroiliac joint inflammation in patients with axial spondyloarthritis (SpA). However, these patients also commonly have active spinal inflammation, and detecting these changes holds diagnostic, prognostic, and therapeutic significance. This study aimed to develop a deep learning algorithm for spinal inflammation based on SPARCC scoring system in patients with axial SpA. Methods: The study cohort included 330 participants with axial SpA. All patients underwent whole spine magnetic resonance imaging (MRI) with short tau inversion recovery (STIR) sequence by 3T MR unit. Three independent readers identified regions of interest (ROIs) to identify bone marrow edema (BME) and performed SPARCC scoring. Two deep learning models based on attention Unet were trained. The BME model was employed to differentiate image with or without spinal inflammation and to delineate BMEs. The vertebral body-intervertebral disc model was utilized to identify discovertebral units. Setting the threshold brightness was applied to detect the region with CSF. The intraclass correlation coefficient (ICC) and Pearson coefficient were used to evaluate the agreement and the correlation between the score of human readers and the score of deep learning-based pipeline. Performance of the models was evaluated using sensitivity, specificity, accuracy, and the Dice coefficient. Results: The ICC and the Pearson coefficient between the SPARCC scores from three human readers and the deep learning-based scoring pipeline were 0.80 and 0.82, respectively. The sensitivity and specificity of identifying image with spinal inflammation were 0.90 and 0.84, respectively. The accuracy of identifying the region containing CSF was 0.88 in images with spinal inflammation. The Dice coefficients were 0.81 (vertebral bodies) and 0.80 (intervertebral disc) in images with spinal inflammation. Conclusion: The high consistency with human readers suggested that the deep learning-based pipeline could offer a SPARCC-informed approach for scoring spinal STIR images in axial SpA.

PURPOSE:To develop and validate DeepMocor, a deep learning-based method for motion-compensated 4-dimensional magnetic resonance fingerprinting (4D-MRF) reconstruction to accelerate conventional 4D-MRF reconstruction, enabling more efficient clinical treatment planning. METHODS AND MATERIALS:This prospective study enrolled 19 hepatocellular carcinoma patients (mean age, 62 years; 14 males) between June 2021 and October 2024. Abdominal free-breathing raw k-space data were acquired using a 3T magnetic resonance imaging scanner. DeepMocor involves motion field initialization, motion field refinement, and final 4D-MRF reconstruction. A 3-fold cross-validation strategy was employed for training and testing. Performance was evaluated against 2 alternatives (stage-I&III-only; stage-III-only) in terms of image quality, tissue property accuracy, tumor-to-tissue contrast, and tumor motion measurement. Image quality was assessed by peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Tissue property accuracy was evaluated by mean absolute percentage error (MAPE). Tumor-to-tissue contrast was quantified by contrast-to-noise ratio (CNR) of the tumor region and the surrounding area. Tumor motion tracking was assessed by average motion difference (AMD) and Pearson correlation coefficients (PCC) in the superior-inferior and anterior-posterior directions. The Wilcoxon signed rank test was used for comparison with P < .05. RESULTS:For T1 maps, DeepMocor demonstrates PSNR of 25.49 ± 1.30, SSIM of 0.84 ± 0.03, MAPE of 3.5% to 5.9%, and CNR of 6.14 ± 3.54. For T2 maps, DeepMocor achieves PSNR of 25.57 ± 1.24, SSIM of 0.88 ± 0.02, MAPE of 3.1% to 15.8%, and CNR of 8.42 ± 13.72. DeepMocor achieves AMD of 0.62 ± 0.86 mm with PCC of 0.96 ± 0.07 in the superior-inferior direction and AMD of 0.32 ± 0.37 mm with PCC of 0.94 ± 0.06 in the anterior-posterior direction. DeepMocor shows superior performance across most metrics compared to stage-III-only and a subset of metrics compared to stage-I&III-only significantly. CONCLUSIONS:The proposed DeepMocor method enables a 24-fold acceleration compared to the conventional reference method, highlighting its potential for liver radiation therapy planning.

Deep learning methods have shown promise in accelerated MRI reconstruction but face significant challenges under domain shifts between training and testing datasets, such as changes in image contrasts, anatomical regions, and acquisition strategies. To address these challenges, we present the first domain generalization framework specifically designed for accelerated MRI reconstruction to robustness across unseen domains. The framework employs progressive strategies to enforce domain invariance, starting with image-level fidelity consistency to ensure robust reconstruction quality across domains, and feature alignment to capture domain-invariant representations. Advancing beyond these foundations, we propose a novel approach enforcing mechanism-level invariance, termed GenCA-MRI, which aligns intrinsic causal relationships within MRI data. We further develop a computational strategy that significantly reduces the complexity of causal alignment, ensuring its feasibility for real-world applications. Extensive experiments validate the framework’s effectiveness, demonstrating both numerical and visual improvements over the baseline algorithm. GenCA-MRI presents the overall best performance, achieving a PSNR improvement up to 2.15 dB on fastMRI and 1.24 dB on IXI dataset at 8× acceleration, with superior performance in preserving anatomical details and mitigating domain-shift problem.

PURPOSE:Recent work has shown MRI is able to measure and quantify signals of phospholipid membrane-bound protons associated with myelin in the human brain. This work seeks to develop an improved technique for characterizing this brain ultrashort- T 2 ∗ $$ {\mathrm{T}}_2\ast $$ component in vivo accounting for T 1 $$ {\mathrm{T}}_1 $$ weighting. METHODS:Data from ultrashort echo time scans from 16 healthy volunteers with variable flip angles (VFA) were collected and fitted into an advanced regression model to quantify signal fraction, relaxation time, and frequency shift of the ultrashort- T 2 ∗ $$ {\mathrm{T}}_2\ast $$ component. RESULTS:The fitted components show intra-subject differences of different white matter structures and significantly elevated ultrashort- T 2 ∗ $$ {\mathrm{T}}_2\ast $$ signal fraction in the corticospinal tracts measured at 0.09 versus 0.06 in other white matter structures and significantly elevated ultrashort- T 2 ∗ $$ {\mathrm{T}}_2\ast $$ frequency shift in the body of the corpus callosum at - $$ - $$ 1.5 versus - $$ - $$ 2.0 ppm in other white matter structures. CONCLUSION:The significantly different measured components and measured T 1 $$ {\mathrm{T}}_1 $$ relaxation time of the ultrashort- T 2 ∗ $$ {\mathrm{T}}_2\ast $$ component suggest that this method is picking up novel signals from phospholipid membrane-bound protons.

Abstract Background The efficacy of levodopa, the most crucial metric for Parkinson’s disease diagnosis and treatment, is traditionally gauged through the levodopa challenge test, which lacks a predictive model. This study aims to probe the predictive power of T1-weighted MRI, the most accessible modality for levodopa response. Methods This retrospective study used two datasets: from the Parkinson’s Progression Markers Initiative (219 records) and the external clinical dataset from Ruijin Hospital (217 records). A novel feature extraction method using MedicalNet, a pre-trained deep learning network, along with three previous approaches was applied. Three machine learning models were trained and tested on the PPMI dataset and included clinical features, imaging features, and their union set, using the area under the curve (AUC) as the metric. The most significant brain regions were visualized. The external clinical dataset was further evaluated using trained models. A paired one-tailed t-test was performed between the two sets; statistical significance was set at p < 0.001. Results For 46 test set records (mean age, 62 ± 9 years, 28 men), MedicalNet-extracted features demonstrated a consistent improvement in all three machine learning models (SVM 0.83 ± 0.01 versus 0.73 ± 0.01, XgBoost 0.80 ± 0.04 versus 0.74 ± 0.02, MLP 0.80 ± 0.03 versus 0.70 ± 0.07, p < 0.001). Both feature sets were validated on the clinical dataset using SVM, where MedicalNet features alone achieved an AUC of 0.64 ± 0.03. Key responsible brain regions were visualized. Conclusion The T1-weighed MRI features were more robust and generalizable than the clinical features in prediction; their combination provided the best results. T1-weighed MRI provided insights on specific regions responsible for levodopa response prediction. Critical relevance statement This study demonstrated that T1w MRI features extracted by a deep learning model have the potential to predict the levodopa response of PD patients and are more robust than widely used clinical information, which might help in determining treatment strategy. Key Points This study investigated the predictive value of T1w features for levodopa response. MedicalNet extractor outperformed all other previously published methods with key region visualization. T1w features are more effective than clinical information in levodopa response prediction. Graphical Abstract

Background The Spondyloarthritis Research Consortium of Canada (SPARCC) scoring system is a sacroiliitis grading system. Purpose To develop a deep learning‐based pipeline for grading sacroiliitis using the SPARCC scoring system. Study Type Prospective. Population The study included 389 participants (42.2‐year‐old, 44.6% female, 317/35/37 for training/validation/testing). A pretrained algorithm was used to differentiate image with/without sacroiliitis. Field Strength/Sequence 3‐T, short tau inversion recovery (STIR) sequence, fast spine echo. Assessment The regions of interest as ground truth for models' training were identified by a rheumatologist (HYC, 10‐year‐experience) and a radiologist (KHL, 6‐year‐experience) using the Assessment of Spondyloarthritis International Society definition of MRI sacroiliitis independently. Another radiologist (YYL, 4.5‐year‐experience) solved the discrepancies. The bone marrow edema (BME) and sacroiliac region models were for segmentation. Frangi‐filter detected vessels used as intense reference. Deep learning pipeline scored using SPARCC scoring system evaluating presence and features of BMEs. A rheumatologist (SCWC, 6‐year‐experience) and a radiologist (VWHL, 14‐year‐experience) scored using the SPARCC scoring system once. The radiologist (YYL) scored twice with 5‐day interval. Statistical Tests Independent samples t ‐tests and Chi‐squared tests were used. Interobserver and intraobserver reliability by intraclass correlation coefficient (ICC) and Pearson coefficient evaluated consistency between readers and the deep learning pipeline. We evaluated the performance using sensitivity, accuracy, positive predictive value, and Dice coefficient. A P ‐value <0.05 was considered statistically significant. Results The ICC and the Pearson coefficient between the SPARCC scores from three readers and the deep learning pipeline were 0.83 and 0.86, respectively. The sensitivity in identifying BME and accuracy of identifying SI joints and blood vessels was 0.83, 0.90, and 0.88, respectively. The dice coefficients were 0.82 (sacrum) and 0.80 (ilium). Data Conclusion The high consistency with human readers indicated that deep learning pipeline may provide a SPARCC‐informed deep learning approach for scoring of STIR images in spondyloarthritis. Evidence Level 1 Technical Efficacy Stage 2

PURPOSE:We aimed to incorporate a deep learning prior with k-space data fidelity for accelerating hyperpolarized carbon-13 MRSI, demonstrated on synthetic cancer datasets. METHODS:A two-site exchange model, derived from the Bloch equation of MR signal evolution, was firstly used in simulating training and testing data, that is, synthetic phantom datasets. Five singular maps generated from each simulated dataset were used to train a deep learning prior, which was then employed with the fidelity term to reconstruct the undersampled MRI k-space data. The proposed method was assessed on synthetic human brain tumor images (N = 33), prostate cancer images (N = 72), and mouse tumor images (N = 58) for three undersampling factors and 2.5% additive Gaussian noise. Furthermore, varied levels of Gaussian noise with SDs of 2.5%, 5%, and 10% were added on synthetic prostate cancer data, and corresponding reconstruction results were evaluated. RESULTS:For quantitative evaluation, peak SNRs were approximately 32 dB, and the accuracy was generally improved for 5 to 8 dB compared with those from compressed sensing with L1-norm regularization or total variation regularization. Reasonable normalized RMS error were obtained. Our method also worked robustly against noise, even on a data with noise SD of 10%. CONCLUSION:The proposed singular value decomposition + iterative deep learning model could be considered as a general framework that extended the application of deep learning MRI reconstruction to metabolic imaging. The morphology of tumors and metabolic images could be measured robustly in six times acceleration using our method.

BackgroundMultiple sclerosis (MS) is a demyelination disease. Myelin water is a biomarker of myelin and thus myelin water imaging is a vital tool to provide insight into the demyelination process.PurposeThis study aimed to characterize the multiple compartments including myelin water fraction (MWF), gray matter (GM) cellular water, white matter (WM) cellular water, and cerebrospinal fluid (CSF) using multiple inversion recovery (mIR) magnetic resonance fingerprinting (MRF) on a clinical MS cohort.MethodsThe Phantom experiment was conducted with tubes containing different WM and GM concentrations extracted from pig brains. For the in-vivo experiment, 23 healthy control (HC) volunteers and 18 MS patients were recruited for this study. The experiments were performed using a clinical 3T MRI. A multi-slice, fast imaging with a steady-state precession (FISP) based mIR MRF protocol was used to obtain the MWF measurements, with 6 min of scan time for each volunteer. The quantification was based on the iterative non-negative least squares (NNLS) with reweighting. The brain compartments quantified were myelin water, WM cellular water, GM cellular water, and CSF. A radiologist with 6 years of experience labeled the MS lesions on FLAIR, MPRAGE, and MWF. Statistical analysis was performed by applying unpaired and paired student's t-tests to compare the MWF results in different groups and in normal-appearing white matter (NAWM) and MS lesions.ResultsThe phantom result demonstrated the ability to detect MWF with various myelin concentrations. The maps derived from mIR MRF, including MWF, WM cellular water, GM cellular water, and CSF were consistent with the anatomical structures observed in FLAIR and MPRAGE. The MWF values in the NAWM of MS patients were significantly different from those in HC, with values of 0.32 +/- 0.025 and 0.25 +/- 0.036, respectively. Additionally, the MWF values in WM lesions were significantly smaller than in NAWM at 0.034 +/- 0.036.ConclusionThe mIR-MRF technique, using multi-compartment analysis, can simultaneously generate maps of MWF, WM cellular water, GM cellular water, and CSF with sufficient brain coverage and in a reasonably short scan time. The MWF map might provide insights into the demyelination associated with MS.