PURPOSE:To develop and externally validate a multimodal stacking fusion model based on dynamic contrast-enhanced MRI (DCE-MRI) for classifying luminal versus non-luminal breast cancer, and to assess each modality's contribution through interpretability and ablation analyses. METHODS:We retrospectively enrolled 396 patients with pathologically confirmed invasive breast cancer from two centers, split into training (n = 220), validation (n = 56), and external test (n = 120) sets. Four unimodal models were trained independently on different feature sources: a clinical multilayer perceptron (Clinical-MLP), a whole-tumor radiomics model (Radiomics-ExtraTrees), a subregion-based habitat radiomics model (Habitat-ExtraTrees), and a transfer learning model (DL-ResNet50). Their predicted probabilities were fused via stacking, with the meta-classifier selected through a grid search over nine candidate model families that identified a support vector machine. Modality contributions were quantified by SHapley Additive exPlanations (SHAP) attribution and leave-one-modality-out ablation experiments. RESULTS:The stacking model achieved AUCs of 0.906, 0.850, and 0.854 on the training, validation, and external test sets, It reached the highest AUC, with significant improvements over Clinical-MLP (P < 0.001) and DL-ResNet50 (P = 0.048), and a sensitivity of 0.839 for identifying luminal cases. SHAP ranked DL-ResNet50 as the dominant contributor (48.6%), and ablation confirmed that its removal caused the largest AUC decline (-0.033). Removing Radiomics marginally increased AUC (+0.003). CONCLUSION:The model achieved robust cross-center performance for luminal versus non-luminal classification. Combining SHAP with ablation revealed that a modality's attribution weight does not guarantee its irreplaceability, suggesting that both analyses are needed to evaluate modality contributions and guide efficient model design.
Major Depressive Disorder (MDD) is a psychiatric disorder associated with altered neuro-metabolism. This study investigated the effect of antidepressant therapy on the metabolism of the right hippocampus and anterior cingulate cortex (ACC) in patients with MDD, using in vivo proton magnetic resonance spectroscopy (MRS) at 3.0 T. MRS was performed in 42 patients with MDD (at baseline and after 8 weeks of antidepressant therapy) and 35 healthy controls (HCs). Metabolite concentrations were quantified using Osprey software with partial-volume corrections. The Hamilton Depression Rating Scale (HDRS) was used to assess response to antidepressant therapy. Patients with MDD who had a > 50% reduction in HDRS scores were classified as remitted, and those with a < 50% reduction were grouped as non-remitted after therapy. Compared with HCs, baseline levels of hippocampal total N-acetylaspartate (tNAA), total creatine (tCr), total choline (tCho), and myo-inositol (mI) were reduced. Baseline levels of ACC tNAA and tCr were higher in patients with MDD than in HCs. After antidepressant therapy, a significant increase in tNAA and tCr levels was observed in the hippocampus of patients with remitted MDD, indicating a positive impact of therapy on energy metabolism and neurodegenerative processes, thereby supporting remission. Hippocampal metabolite levels remained unaltered after therapy in the non-remitted subgroup. ACC metabolites remained unaltered after therapy in both remitted and non-remitted subgroups. Thus, findings highlighted the importance of hippocampal metabolism in MDD and the possibility that metabolite levels may serve as biomarkers of antidepressant therapy response.
BACKGROUND/PURPOSE:Simultaneous T1/T2 mapping with hybrid (mixed) sequences has been proposed, but clinical use is limited by long acquisition times. Accelerated reconstruction-parallel imaging (PI; SENSE), compressed sensing (CS), and deep learning (DL)-allows shorter scans but may alter quantitative values and image appearance. This phantom study compared SENSE, CS, and DL across acceleration factors and examined how quantitative deviation relates to image-similarity metrics. METHODS:Using non-PI reference T1/T2 maps, we reconstructed hybrid maps with SENSE, CS, and DL at nominal reduction factors of 2-10. Circular ROIs were placed at the centers of nine phantom rods on the reference T1 map and identically propagated. Absolute percentage deviation of mean ROI T1/T2 values quantified agreement with the reference; PSNR and SSIM quantified similarity to the reference. Between-method differences at each reduction factor were tested with Friedman and, when significant, Bonferroni-corrected Wilcoxon signed-rank tests. RESULTS:Reference T1/T2 values ranged from 527 to 2138 ms and 55-572 ms, respectively. Acceleration effects were method- and metric-dependent: SENSE deviations increased at higher acceleration, whereas CS and DL preserved smaller deviations. At 10×, median T1/T2 deviations were 27.57%/31.47% for SENSE, versus 1.55%/1.40% for CS and 4.24%/2.15% for DL. Image-similarity rankings were not consistently concordant with quantitative-deviation rankings, and several factor-metric combinations showed significant between-method differences (p < 0.05). CONCLUSION:In accelerated hybrid T1/T2 mapping, quantitative deviation and PSNR/SSIM can diverge, so single-metric evaluation may be insufficient. A combined framework using quantitative deviation, PSNR/SSIM, and representative images can prioritize candidate acceleration and reconstruction conditions for subsequent task-specific in vivo validation, but does not by itself establish clinical acceptability.
OBJECTIVES:To evaluate the performance of nomograms combining clinical factors, apparent diffusion coefficient (ADC), and radiomics features from functional MRI parametric maps in predicting deep myometrial invasion (DMI), high histopathological grade, and lymphovascular space invasion (LVSI) in early endometrial cancer (EC). METHODS:This multicenter study recruited 362 patients with EC undergoing preoperative MRI. Radiomics features were extracted from four functional MRI parametric maps and selected via a three-step process. High-risk factors were identified by logistic regression. Nonradiomics models (clinical high-risk factors plus ADC), radiomics models (radiomics features), and nomogram models combining both were developed and validated for predicting DMI, high histopathological grade, and LVSI. Performance was assessed by receiver operating characteristic analysis. RESULTS:Logistic regression identified age and ADC as predictors of DMI, tumor size and ADC of high-grade lesions, and CA125 and ADC of LVSI; 10, 13, and 12 radiomics features constituted the respective Radscores. In the external testing set, the areas under the curve (AUCs) of the nonradiomics, radiomics, and nomogram models were 0.691 (95% confidence interval [CI]: 0.578-0.804), 0.756 (95% CI: 0.665-0.848), and 0.785 (95% CI: 0.709-0.849) for predicting DMI; 0.718 (95% CI: 0.613-0.823), 0.876 (95% CI: 0.796-0.956), and 0.906 (95% CI: 0.846-0.948) for high histopathological grade; and 0.651 (95% CI: 0.545-0.756), 0.764 (95% CI: 0.653-0.875), and 0.808 (95% CI: 0.716-0.900) for LVSI. CONCLUSION:The nomogram combining radiomics features from functional MRI parametric maps with clinical factors and ADC performs promisingly in evaluating DMI, high histopathological grade, and LVSI in early EC, demonstrating potential for personalized management.
Magnetic resonance spectroscopy (MRS) provides non-invasive in vivo assessment of brain tumour metabolism and complements conventional magnetic resonance imaging (MRI) by adding biochemical information beyond structural imaging. This narrative state-of-the-art review summarises current clinical and emerging applications of proton MRS and magnetic resonance spectroscopic imaging (MRSI) in brain tumours, with emphasis on metabolic biomarkers, technical developments, clinical translation, and current limitations. Established spectroscopic patterns, particularly elevated choline, reduced N-acetylaspartate, and variable lactate and lipid signals, support lesion characterisation, glioma grading, and differentiation of tumour recurrence from treatment-related effects. MRSI further enables spatial mapping of metabolic heterogeneity, helping to identify infiltrative and non-enhancing tumour regions relevant to biopsy targeting, surgical planning, and radiotherapy guidance. Molecularly informative metabolites, especially 2-hydroxyglutarate in isocitrate dehydrogenase-mutant gliomas, extend the role of MRS toward non-invasive molecular imaging, while emerging markers such as alanine and glycine remain promising but require further validation. Advances in localisation, acceleration, spectral editing, automated quantification, and artificial intelligence-assisted analysis have improved feasibility and reproducibility, although variability in acquisition protocols, quantification methods, quality control, and reporting continues to limit widespread clinical adoption. Hyperpolarized 13C and deuterium-based methods offer dynamic assessment of tumour metabolism, but largely remain translational or research-oriented techniques. Overall, MRS and MRSI are increasingly positioned as useful components of multiparametric neuro-oncology imaging, provided that future work continues to prioritise protocol harmonisation, multicentre validation, clinically actionable thresholds, and integration into routine diagnostic and treatment-planning workflows.
Magnetic resonance imaging (MRI) is the preferred modality for non-invasive examination of the human brain, but the acquired images are inevitably corrupted by Rician noise and by intensity non-uniformity (INU), also called the bias field, which arises from imperfect radio-frequency coils and from subject-induced field perturbations, and these artefacts degrade subsequent registration, segmentation and tumour-detection pipelines. This article reviews 151 publications spanning four decades (1985 to early 2026) on the removal of noise and intensity inhomogeneity from brain MRI, located through a structured search of IEEE Xplore, PubMed, Scopus, Web of Science and Google Scholar and complemented by manual snowballing. The methods surveyed are organised into prospective approaches (phantom-based, multi-coil and special-sequence techniques), retrospective approaches (filtering, surface fitting, segmentation-based and histogram-based correction), classical machine learning, and deep learning (convolutional networks, generative adversarial networks, denoising autoencoders, transformer-based architectures, denoising diffusion probabilistic models and self-supervised frameworks). Deep-learning approaches to bias-field correction are reviewed separately and classified into direct field regression, adversarial translation, joint estimation of the field and the uniform image, and frequency-domain probabilistic formulations, a distinction that clarifies why supervision, rather than architecture, is the limiting factor in that literature. On the BrainWeb T1 benchmark at Rician noise σ = 9%, 3D-Parallel-RicianNet reaches a published peak signal-to-noise ratio (PSNR) of 37.11 dB and structural similarity index (SSIM) of 0.9859, compared with 28.91 dB and 0.9584 for BM3D. The 2023-2026 generation of models-Swin-UNet Transformers, the Imaging Transformer for SNR ≪ 1, hybrid HTC-Net, the lightweight CHARMS CNN-Transformer, denoising diffusion probabilistic models and self-supervised frameworks such as Coil2Coil, SURE-Net, Rep2Rep and the Efficient Collaborative Diffusion Model-report further methodological gains and remove the requirement for paired noise-free training data, which had been the principal obstacle to clinical deployment. Open challenges remain in cross-scanner generalisation, in clinically meaningful evaluation beyond PSNR and SSIM, and in the construction of a publicly available multi-vendor benchmark for brain-MRI denoising.
Low-field magnetic resonance imaging (MRI) is an affordable medical imaging technique used to assess the structural and functional features of internal and external tissues and organs. However, its clinical effectiveness is often constrained by prolonged scan times and reduced image quality due to compromised signal-to-noise ratios. Deep learning (DL) has emerged as an effective solution for reconstruction of undersampled MRI data. This study presents a comprehensive comparative evaluation of six U-Net variants, including U-Net, Attention U-Net, Residual U-Net, Recurrent-Residual (R2) U-Net, Residual-Attention U-Net (RAU-Net), and U-Net++, for image domain reconstruction of low-field (0.3 Tesla) MRI data from the M4RAW dataset. Root Sum of Squares (RSS) magnitude images were used for training and testing. The models were trained on 1024 MRI volumes comprising 18,432 image slices. Reconstruction performance was evaluated under Cartesian undersampling at acceleration factors of 2, 4, 8, and 16, and estimated different statistical indices including Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), Normalized Mean Squared Error (NMSE) and Coefficient of Determination (R2). The results demonstrate that the vanilla U-Net performs well at lower acceleration factors, achieving average SSIM and PSNR values of 0.9303 and 39.18 dB, respectively, at an acceleration factor of 2. In contrast, more complex architectures, particularly U-Net++ and RAU-Net, exhibit superior performance at higher acceleration factors. In particular, U-Net++ achieved average SSIM values of 0.8754 and 0.8731, and PSNR values of 35.21 dB and 35.06 dB at acceleration factors of 8 and 16, respectively. These findings inform the development of efficient and accessible imaging solutions, promote the broader adoption of low-field MRI, and encourage further innovation in cost-effective medical imaging technologies.
BACKGROUND AND OBJECTIVES:Hypertrophic Cardiomyopathy (HCM) is the most prevalent inherited cardiomyopathy. Its left ventricular hypertrophy (LVH) phenotype exhibits substantial radiological overlap with HCM phenocopies such as Hypertensive Heart Disease (HHD) and Cardiac Amyloidosis (CA), posing significant challenges to clinical differential diagnosis. Differentiating HCM from its phenocopies and normal hearts on cardiac MRI is an important but difficult task. This study leveraged a cardiovascular magnetic resonance foundation model to develop a deep learning approach that enables accurate and automated differentiation among HCM, normal controls, and HCM phenocopies. METHODS:We integrated 463 cases from a multi-center dataset comprising two public benchmarks, ACDC and M&Ms2, along with a clinical cohort from a tertiary medical center. The final cohort included 207 HCM cases, 166 normal controls, and 90 cases of HCM phenocopies (including HHD, CA, and aortic stenosis, among other etiologies). To address the challenge of differential diagnosis in HCM, a tailored network was developed based on the Convolution-Transformer hybrid backbone of the CineMA cardiac MRI foundation model. The architecture employs a dual-head multi-task learning design: a primary classification head performs three-class diagnosis (HCM vs. Normal vs. Other), while an auxiliary classification head targets fine-grained discrimination between HCM and non-HCM myocardial hypertrophy. A Boundary Prior Attention Module (BPAM) was further integrated into the convolutional encoder, utilizing left ventricular myocardial segmentation masks as anatomical priors to guide spatially selective feature modulation. A three-stage progressive transfer learning strategy was implemented, transitioning from frozen backbone to partial unfreezing and finally to global optimization, with class-weighted cross-entropy and Focal Loss to address class imbalance. RESULTS:The proposed method achieved an accuracy of 85.6% ± 1.7%, a macro-averaged F1-score of 0.840 ± 0.019, and an ROC AUC of 0.954 ± 0.008 under five-fold cross-validation, significantly outperforming the randomly initialized baseline (F1-score 0.704 ± 0.021, AUC 0.853 ± 0.014). HCM detection achieved a recall of 85.5% ± 2.4%, indicating a low rate of missed diagnoses in this high-risk population. Ablation experiments demonstrated that pretrained weights, the dual-head architecture, the progressive training strategy, and BPAM contributed incremental AUC gains of +6.5%, +1.2%, +1.1%, and + 1.3%, respectively. CONCLUSIONS:The CineMA-based framework incorporating dual-head multi-task learning, three-stage progressive transfer learning, and anatomy-guided boundary prior attention achieves robust and accurate differentiation among HCM, normal controls, and HCM phenocopies. The approach demonstrates promising potential for AI-assisted HCM screening and clinical decision support across multi-center, multi-device settings.
PURPOSE:Quantitative multiparametric MRI (mpMRI) is increasingly explored for clinically significant prostate cancer (csPCa) detection; however, the diagnostic value of quantitative dynamic contrast-enhanced (DCE)-MRI in prostate cancer remains controversial. In this study, we aim to evaluate the diagnostic performance of water-exchange DCE-MRI in distinguishing prostate tumors from benign tissue and differentiating csPCa from clinically insignificant (CIS) disease. METHODS:This retrospective study included 89 patients who underwent prostate DCE-MRI between March 2022 and October 2023. DCE-MRI quantitative analysis was based on two-site water exchange (2SX) and conventional Tofts model. Tumor, benign and benign prostatic hyperplasia regions of interest were drawn based on quantitative multiparametric MRI. Student's t-test was implemented for intergroup comparisons between benign tissue and tumor, and between CIS and csPCa. Diagnostic performance for differentiating tumor from benign tissue and csPCa from CIS was accessed via linear discriminant analysis with tenfold cross-validation. RESULTS:Fifty-nine patients (mean age ± standard deviation, 69 ± 8 years) were finally included. Most MRI voxels in tumor (86%) and benign tissue (76%) favored 2SX model over Tofts model based on corrected Akaike's Information Criterion. The volume transfer constant (Ktrans) and cellular water efflux rate constant (kio) from 2SX model were significantly higher in tumor regions than benign regions (both P < 0.0001). Compared with the Tofts model, the 2SX model improved discrimination between tumor and benign tissue. Moreover, incorporating kio to correct Ktrans, 2SX model significantly enhanced the ability of Ktrans to differentiate csPCa from CIS compared with Tofts model (area under curve: 0.72 vs 0.63). CONCLUSION:The 2SX model incorporating transmembrane water exchange into quantitative DCE-MRI analysis improves diagnostic performance over the conventional Tofts model alone for distinguishing prostate tumor from benign tissue and for differentiating csPCa from CIS.
PURPOSE:To evaluate diffusion-weighted imaging (DWI) and MR Elastography (MRE) performance over a clinically relevant stiffness range for liver fibrosis with and without the presence of compressive motion using an in vitro phantom pipeline. METHODS:Five anthropomorphic liver phantoms were created using polyacrylamide (PAA) hydrogels. Phantom stiffness ranged from approximately 1.5-11 kPa. Phantoms were connected to a pulsatile flow circuit to induce compressive motion. Imaging was performed on a 3T MRI system. MRE, conventional DWI, and M1-motion optimized DWI (MODI) were assessed at varying pulsatile flow states (mean flow = 0, 0.5, 1, 1.5 L/min). Conventional and MODI DWI apparent diffusion coefficients (ADC) were correlated to MRE stiffness. RESULTS:Stiffness and ADC maps were obtained for all phantoms and flow states. Conventional and MODI DWI under static conditions exhibited an inverse relationship with MRE stiffness (R2 = 0.94 and R2 = 0.95, respectively). MODI-DWI exhibited reduced motion-induced signal dropout, minimizing ADC bias under motion conditions compared to conventional DWI. Mean MRE stiffness was largely unaffected by motion, except for the lowest and highest stiffness phantoms at the highest flow rate. CONCLUSION:Anthropomorphic liver phantoms were used to systemically investigate the effects of stiffness and motion on the performance of MRE and DWI. MRE and MODI-DWI demonstrated strong insensitivity to motion across stiffness and motion level.
BACKGROUND AND PURPOSE:Central nervous system (CNS) relapse in diffuse large B-cell lymphoma (DLBCL) represents a clinical challenge, with risk stratification based on the CNS International Prognostic Index (CNS-IPI) potentially leading to over- and undertreatment. This study investigated whether diffusion MRI metrics could detect CNS microstructural changes associated with CNS relapse risk factors. MATERIALS AND METHODS:In this cross-sectional study 63 DLBCL patients without CNS involvement, and 11 healthy controls, underwent brain MRI including multi-shell diffusion tensor imaging, utilising restriction spectrum imaging modelling. We analysed cellularity index (CI), fractional anisotropy (FA), mean diffusivity (MD) and neurite density (ND) in white matter (WM) and CI and MD in cortical grey matter (GM). Associations with CNS-IPI, risk factors, and blood-brain barrier (BBB) permeability were assessed. RESULTS:After multiple comparisons age-adjusted linear regression showed higher GM MD in patients with high CNS-IPI versus low/intermediate (mean difference: 4.33 × 10-5, FDR-p = 0.03). Compared with controls, patients had increased GM MD (mean difference: 3.93 × 10-5 mm2/s, FDR-p = 0.003) and reduced WM ND (mean difference: -7.36 × 10-5, FDR-p = 0.003). Kidney/adrenal involvement (mean difference: 4.48 × 10-5 mm2/s, FDR-p = 0.002) and advanced stage (mean difference: 3.41 × 10-5 mm2/s, FDR-p = 0.02) were associated with higher GM MD. MD correlated positively with BBB permeability in GM (ρ = 0.46) and WM (ρ = 0.26), whereas WM FA (ρ = -0.37) and CI (ρ = -0.32) correlated inversely all p < 0.05. CONCLUSIONS:DLBCL patients demonstrate subtle CNS microstructural alterations detectable by diffusion MRI, particularly in high-risk groups, suggesting potential brain vulnerability preceding clinical CNS involvement.
Q-space trajectory imaging (QTI) provides promising markers of tissue microstructure, but clinical translation requires shorter acquisitions, faster analysis, and more robust parameter estimation at high spatial resolution. To address these barriers, we trained a voxel-wise multilayer perceptron (MLP) to infer QTI-derived scalar parameters directly from the diffusion signal. We established reference QTI parameters of the brain in 18 healthy subjects using constrained fitting on 50-min QTI scans. The MLP was trained to estimate those reference parameters from a five-minute subset of the diffusion data. We compared the MLP with the constrained fit applied to the same short-protocol input, computing normalized root mean squared error, peak signal-to-noise ratio, and structural similarity with respect to the reference. Here, the MLP consistently achieved better performance metrics, with normalized root mean squared errors up to two-fold lower. For one whole-brain dataset, MLP inference reduced computation time from more than an hour with constrained fitting to a few seconds. Robustness to lower SNR was tested in a separate 1.7 mm isotropic voxel size acquisition of the full protocol, in which the MLP retained lower errors and less visually apparent noise. Finally, we demonstrate qualitative feasibility in two glioma patients scanned with the short protocol. We conclude that a simple MLP can provide high-quality QTI parameter estimates from short tensor-valued diffusion acquisitions. This enables five-minute, high-resolution QTI and may encourage further clinical studies of markers such as microscopic fractional anisotropy.
Mechanistic modelling of the blood oxygenation level-dependent (BOLD) response is essential for interpreting functional magnetic resonance imaging (fMRI) in physiological terms. A central framework in this literature is the balloon model, which links changes in cerebral blood flow, venous blood volume, and deoxyhemoglobin content to the observed BOLD signal through a compliant venous compartment. This review examines the development of models that extend or embed this balloon/Windkessel lineage. We define the balloon framework as hemodynamic models that retain a compliant venous or Windkessel-like compartment, represent blood volume and deoxyhemoglobin as latent dynamical states, and generate BOLD contrast through balloon-type vascular behaviour. The definition presented intentionally distinguishes the balloon family from a broader set of compliant-compartment hemodynamic models. Within this scope, we review classical balloon variants, extensions involving viscoelasticity, autoregulation, oxygen-extraction assumptions, and multicompartment physiology, anatomically structured adaptations for laminar fMRI and white-matter BOLD modelling, and broader generative formulations including dynamic causal modelling, stochastic state-space models, and multimodal electroencephalogram-fMRI (EEG-fMRI) frameworks. We also consider studies of validation, sensitivity, and parameter identifiability. Across this literature, the balloon model emerges not as a single fixed formulation but as a flexible family of mass-balance-based hemodynamic models. Its continued value lies in its physiological interpretability, modularity, and compatibility with inference frameworks, while its main limitations remain limited anatomical specificity, uncertainty in neurovascular coupling, and limited identifiability from BOLD data alone.
Gadolinium-based contrast agents (GBCAs) are essential for the evaluation of central nervous system (CNS) disorders, enhancing lesion detection, characterization, and therapeutic monitoring. Despite their widespread use and overall safety, their application in daily practice may be, in some cases, suboptimal due to technical or interpretative issues, including incorrect dosing, inadequate timing of acquisition, and sequence parameter adjustments. Although multiple GBCAs demonstrate comparable diagnostic performance, safety concerns such as hypersensitivity reactions, nephrogenic systemic fibrosis, and gadolinium retention have led to increasing emphasis on optimized and justified use. Regulatory differences between agencies, alongside growing preference for macrocyclic agents, reflect efforts to balance diagnostic benefit and safety. From a physicochemical standpoint, relaxivity, molecular structure, and pharmacokinetics critically influence imaging performance, particularly through T1 signal enhancement. Recent advances support dose reduction strategies, especially with high-relaxivity agents, without compromising diagnostic accuracy. Technical factors, including magnetic field strength, acquisition timing, and sequence selection, further modulate enhancement quality. Beyond conventional imaging, advanced techniques such as perfusion MRI (Dynamic Susceptibility Contrast and Dynamic Contrast Enhanced), Dixon imaging, black-blood sequences, and susceptibility-weighted imaging provide complementary physiological and structural information, enhancing diagnostic precision. These approaches enable more efficient use of GBCAs and increase diagnostic accuracy. Moreover, emerging artificial intelligence applications may further transform contrast utilization by enabling high-quality imaging at lower doses. In conclusion, a comprehensive, technically optimized, and physiologically informed approach to GBCA use is crucial to maximizing diagnostic yield while minimizing potential risks in CNS MRI.
Cardiac T1 mapping is susceptible to respiratory motion, particularly due to the substantial contrast variations and signal inversions across different inversion times (TI). This study proposes a novel training-free motion correction framework leveraging frozen DINOv3 foundation model features to achieve robust myocardial alignment without task-specific network training or fine-tuning. For each image pair, dense DINOv3 features extracted via a frozen ViT-S/16+ encoder are projected into a compact pair-wise PCA feature space. We introduce a variance-aware channel selection and adaptive weighting strategy to prioritize structurally reliable features across the TI-dependent contrast fluctuations. The deformation field is optimized through a hybrid loss function combining weighted DINO-based normalized cross-correlation (NCC), an auxiliary image-domain NCC constraint, and smoothness regularization. Evaluated on the public STONE T1 mapping dataset (32 subjects, 1600 image pairs), the proposed method outperforms existing image-based registration baselines, achieving a Dice similarity coefficient (DSC) of 0.839 ± 0.090 and an HD95 of 1.965 ± 1.705. Furthermore, the method yields the highest T1 fitting quality, with a myocardial R2 of 0.982 ± 0.016, while maintaining topology preservation. Our framework provides solution for cardiac T1 mapping, improving structural alignment under intensity variations without requiring large-scale annotated training data.
High-field magnetic resonance imaging (MRI) systems impose stringent performance requirements on gradient coils, including high efficiency, excellent field linearity, and sufficient mechanical robustness. In this work, an actively shielded and force-balanced gradient coil assembly was developed for a cryogen-free 7 T/160 mm superconducting MRI magnet. The gradient coils were designed using a conventional boundary element method (BEM)-based framework with active shielding, force-balance constraints, and engineering manufacturability considerations to improve gradient efficiency and suppress electromagnetic forces under strong background magnetic fields. Based on the optimized design, three-axis gradient coils were fabricated using CNC machining and axial winding and integrated into a compact coaxial multilayer assembly through a standardized engineering fabrication and assembly process. Experimental results show that the X-, Y-, and Z-axis coils achieve gradient efficiencies of 3.95 mT/m/A, 3.72 mT/m/A, and 4.12 mT/m/A, respectively, with field linearity errors within ±2.5% inside a 50 mm diameter spherical volume (DSV). The measured electrical parameters show good agreement with the design values, confirming the feasibility of the proposed fabrication and assembly method. These results demonstrate the engineering feasibility and practical implementation capability of the proposed gradient coil assembly for compact high-field micro-MRI applications.
OBJECTIVE:To develop and externally validate an integrated model that combines multimodality CT-MRI deep learning with clinical and radiological features for noninvasive preoperative hepatocellular carcinoma (HCC) histologic grading, and to evaluate whether this integration outperforms single-modality imaging, the multimodality imaging model alone, and a clinical-radiological model. METHODS:In this multicentre retrospective study, 668 patients with pathologically confirmed HCC from three institutions (January 2010-December 2023) were included. Single-modality cohorts from Centre 1 (CT-only, n=283; MRI-only, n=135) were used for modality-specific pretraining via staged transfer learning. A total of 250 patients with paired preoperative CT and MRI were allocated to a training cohort (n=88), an internal validation cohort (n=57), and two independent external validation cohort (n=62, n=43). Single-modality multiphase CT (mpCT) and multisequence MRI (msMRI) models, a combined multimodality CT-MRI deep learning model (TL-CMDLM), a clinical-radiological signature model (CRSM), and an integrated hybrid fusion model (TL-HFM) combining the CT-MRI deep learning signature with clinical and radiological features were developed. Performance was assessed using the area under the receiver operating characteristic curve (AUC), integrated discrimination improvement, and decision curve analysis. RESULTS:A total of 250 patients with paired CT and MRI (218 men; mean age, 57 ± 10 years) were evaluated. The integrated TL-HFM achieved the highest discrimination, with AUCs of 0.89, 0.88, and 0.90 in the internal validation cohort and two external validation cohort, respectively, and significantly outperformed both the multimodality CT-MRI deep learning model alone (TL-CMDLM: 0.82, 0.72, 0.81) and the clinical-radiological model (CRSM: 0.62, 0.73, 0.59) across all cohorts (all p < 0.05). The multimodality CT-MRI model in turn outperformed each best single-modality model (mpCT: 0.76, 0.58, 0.61; msMRI: 0.72, 0.62, 0.62). Decision curve analysis confirmed that the integrated TL-HFM provided the greatest net benefit across the internal validation and two external validation cohort. CONCLUSION:An integrated model combining multimodality CT-MRI deep learning with clinical and radiological features provided the most accurate noninvasive preoperative grading of HCC, outperforming single-modality imaging, multimodality imaging alone, and a clinical-radiological model, and delivered the greatest net clinical benefit across multicentre cohorts.
PURPOSE:Quantitative MRI biomarkers are increasingly used to assess liver health; however, athlete-specific normative values are lacking. This study aimed to establish liver-specific normative values for proton density fat fraction (PDFF), R2*-based liver iron concentration (LIC), and T1 and T2 relaxation times in collegiate athletes. METHODS:In this retrospective single-center study, collegiate athletes undergoing return-to-play cardiac MRI (1.5 T/3.0 T) following Coronavirus Disease 2019 recovery were identified. Whole-liver PDFF and R2* were obtained from chemical shift-encoded MRI, T1 and T2 relaxation times were derived from cardiac mapping with partial liver coverage. LIC was calculated from R2*. PDFF and LIC were considered field-strength-independent, whereas R2*, T1, and T2 were analyzed by field strength. Biomarkers are reported as median (interquartile range). RESULTS:A total of 211 athletes (64.0% males; median age 19.9 years) were included. PDFF and LIC were 2.2% (1.7, 2.8) and 0.67 mg/g (0.58, 0.74) in males and 1.8% (1.5, 2.3) and 0.57 mg/g (0.50, 0.67) in females. At 1.5 T, R2*, T1 and T2 were 34.0 s-1 (31.9, 36.4), 571 ms (552, 593), and 45.4 ms (43.3, 47.0) in males and 32.0 s-1 (29.5, 34.6), 588 ms (577, 600), and 49.4 ms (46.4, 52.2) in females. At 3.0 T, values were 44.1 s-1 (41.8, 48.8), 755 ms (727, 783), and 40.9 ms (38.3, 43.4) in males and 41.6 s-1 (37.7, 44.0), 806 ms (791, 822), and 42.8 ms (38.9, 44.3) in females. CONCLUSIONS:We established sex- and field strength-specific normative values for quantitative liver MRI biomarkers in collegiate athletes, providing a foundation for baseline clinical interpretation and longitudinal monitoring of liver health.
BACKGROUND:Cardiac magnetic resonance elastography (MRE) is an emerging modality for noninvasive assessment of left ventricular (LV) myocardial stiffness. Accurate LV myocardium delineation is essential for MRE analysis, yet current workflows often rely on manual annotation and additional structural MRI. It remains uncertain whether native cardiac MRE data alone are sufficient for reliable automated LV segmentation. PURPOSE:To evaluate deep learning approaches for LV myocardium segmentation on cardiac MRE data and to assess the influence of input representation and automation strategy on segmentation performance. METHODS:Cardiac MRE data from 16 healthy male volunteers were used to train and evaluate two contemporary segmentation frameworks, nnU-Net v2 and MedSAM. Reader 1 annotated the full dataset using MRE magnitude images, and Reader 2 independently annotated the test set, enabling model performance to be benchmarked against inter-reader agreement. nnU-Net was trained using multiple input representations and training strategies. MedSAM was evaluated in zero-shot, semi-automated, fine-tuned, autoprompt, and box-regression configurations. RESULTS:Inter-reader Dice agreement was 0.79 ± 0.03. The best nnU-Net model, trained on fully averaged normalized magnitude images, achieved a Dice score of 0.82 ± 0.04. Performance was lower with magnitude-plus-phase and real-plus-imaginary inputs, with Dice scores of 0.65 ± 0.21 and 0.60 ± 0.20, respectively, and also decreased with non-normalized magnitude input, which yielded a Dice score of 0.75 ± 0.05. The best MedSAM result was obtained with a semi-automated fine-tuned variant using strong ROI smoothing, which achieved a Dice score of 0.82 ± 0.02. Fully automated MedSAM variants performed less well, with Dice scores of 0.68 ± 0.09 for autoprompt and 0.71 ± 0.08 for box regression. CONCLUSIONS:Cardiac MRE data alone demonstrated the feasibility of accurate LV myocardium segmentation, with nnU-Net and MedSAM both reaching inter-reader-level performance. These findings support direct segmentation of the LV myocardium from native cardiac MRE and represent a step toward a self-contained cardiac MRE workflow.
Pathological iron accumulation is a common pathophysiological hallmark across multiple neurodegenerative diseases (NDDs), motivating the need for accurate, non-invasive quantification methods. Quantitative susceptibility mapping (QSM) is an advanced magnetic resonance imaging (MRI) technique that enables in vivo measurement of tissue magnetic susceptibility (χ), providing a sensitive proxy for iron content. This umbrella review systematically evaluates the diagnostic accuracy, clinical correlations, and distinct iron distribution patterns of QSM in major NDDs, such as Parkinson's disease (PD), Alzheimer's disease (AD), amyotrophic lateral sclerosis (ALS), and atypical Parkinsonism. We included 15 (13/15 were rated Low or Critically Low on AMSTAR 2) systematic reviews and meta-analyses (through July 15, 2026); however, the findings should be interpreted cautiously because of heterogeneity and the low methodological quality. A Corrected Covered Area (CCA) analysis demonstrated only slight overlap of primary studies across the included reviews (CCA = 5.42%). Collectively, the evidence indicates that QSM provides comparable or higher diagnostic sensitivity and reliability than conventional R2* and SWI techniques, particularly for deep gray matter structures. The findings support significant iron overload in the substantia nigra, particularly in the pars compacta, as a robust biomarker for PD that correlates with motor severity and disease duration. Furthermore, regional iron profiling in the basal ganglia is critical for differential diagnosis; specifically, elevated χ in the putamen and globus pallidus effectively distinguishes multiple system atrophy and progressive supranuclear palsy from idiopathic PD. Distinctively, AD and ALS exhibit specific χ alterations in the thalamus, motor cortex, and hippocampus, reflecting divergent iron-related pathophysiological mechanisms, which correlate with cognitive impairment and upper motor neuron signs. Overall, QSM shows diagnostic promise and offers mechanistic insights into iron-related neurodegenerative processes.
