Impact: While IQMs are commonly used to determine the superiority of image acquisition or reconstruction methods, differing conclusions about whether noise correlation decreases quality exemplifies why one should be cautious when using IQMs as indicators of technical performance.

Impact: This work introduces a GRAPPA-based water–fat separation for single-shot diffusion EPI, eliminating the need for fat-saturation or multi-echo acquisitions. It improves fat suppression efficiency and reduces scan time, potentially enhancing diffusion MRI robustness across anatomies.

Impact: The proposed zero-shot self-supervised reconstruction with separate learned regularization for magnitude and phase ensures robust elastogram estimation from accelerated acquisitions without training data. This development facilitates rapid MRE, enhancing clinical feasibility and patient throughput in routine practice.

Impact: UNITS grounds diverse empirical self-supervised MRI reconstruction methods in theory and unifies them under a single framework. This foundation establishes guiding principles for future method design and opens a new self-supervised learning era for robust, reference-free, and generalizable MRI reconstruction.

Impact: 
We deliver rapid voxelwise noise and g-factor maps for iterative and compressed-sensing MRI, replacing slow Pseudo-Multiple-Replica simulations. Our unbiased stochastic estimator enables quantitative SNR assessment, and can inform sequence/reconstruction tuning and on-the-fly, and adaptive protocol optimization toward practical, noise-aware MRI.

Impact: Generative AI may learn reproducible distributions of the MRI space that can be guided towards efficient multi-sequence reconstructions, significantly reducing overall protocol scan time.

Impact: ReconHMT is a novel physics-guided transformer that reconstructs MR images within the hierarchical latent space of a foundational autoencoder model. By leveraging adapters to enforce token-level coherence and data-consistency across hierarchical stages, ReconHMT achieves high-fidelity reconstructions while maintaining computational efficiency.

Impact: Enables faster, high-resolution volumetric T2w brain MRI without t2 blurring, improving depiction of subtle pathology. Motivates diagnostic performance in patient populations.

Impact: We proposed HiFi-QSM, a novel framework for high-resolution QSM reconstruction. The proposed method is expected to enable QSM for high-resolution (<< 1 mm) images, which is beneficial for visualizing fine brain structures both in vivo and ex vivo.

Impact: PyGrog improves non-Cartesian MRI reconstruction by providing an efficient, open-source GROG-based alternative to NUFFT. Its speed, interoperability, and accuracy support practical integration into modern imaging workflows, enhancing accessibility and performance for research and clinical MRI applications.

Impact: This method significantly reduced reconstruction errors from head motion, providing improved cortical thickness estimates. In addition, the improved image quality may benefit clinical imaging studies without the need for sedation.

Impact: This framework enables real-time, continuous 3D volumetric motion tracking for MRIgRT from highly undersampled k-space, without requiring ground-truth training data. It can capture complex, multi-dimensional motion patterns, paving the way for more accurate tumour tracking and organ-at-risk avoidance during radiotherapy.

Impact: Joint motion identification, reconstruction and fitting enable efficient multi-dose contrast-enhanced myocardial T₁ mapping from only 1-minute scans per dose, even in 3D, offering a potential path toward quantitative multi-dose imaging with improved accuracy and reduced scan time.

Impact: DeepGuidedGS provides a practical approach for utilizing prior scans in clinical follow-ups, supporting faster imaging and improved reconstruction quality even when anatomy changes or alignment is imperfect. This may enhance efficiency and reliability in longitudinal monitoring workflows.

Impact: We proposed MoDL reconstruction with reference-guided U-Net for self-navigated simultaneous multi-slab 3D DWI, which achieved a >50-fold acceleration in total reconstruction time without compromising the image quality. This facilitates robust, high-resolution diffusion imaging without the barrier of prohibitive computation time.

Impact: This feasibility study demonstrates low-field 0.55T UTE MRI with diffusion model-based deep learning reconstruction enabling high-resolution, radiation-free craniofacial imaging. Supporting radiation free orthodontic workflows, reproducible cephalometric analysis, and improved visualization of bone and soft tissue for clinical and research use.

Impact: High-resolution TOF-MRA with deep-learning-reconstruction can significantly raise small vessels visibility. This imaging technique could serve as a valuable noninvasive tool for assessment of degree of collateral circulation in moyamoya disease, and others such as lenticulostriate arteries and aneurysms.

Impact: The BUDA-iQSM+ framework, with orientation-adaptive latent feature editing, enables fast, artifact-free, high-resolution whole-brain QSM. This innovation advances neuroscience research and clinical applications by providing reliable susceptibility quantification, targeting MR physicists, radiologists, and scientific professionals engaged in neuroimaging and diagnostic studies.