Impact: Highly accelerated multi-shot EPIK MRI, integrated with deep learning–based distortion correction, enables rapid and distortion-suppressed high-resolution whole-brain imaging. This approach enhances spatial fidelity and supports improved signal quality compared to conventional EPIK, facilitating more precise functional mapping at ultra-high fields.

Impact: Our approach uses pseudo-random sampling and compressed sensing to generate high quality fat–water cine MRI within feasible breath-holds, with the aim of enhancing detection of intramyocardial fatty infiltration and pericardial abnormalities, while reducing scan time and improving patient tolerance.

Impact: Compressed sensing accelerates acquisition, whereas higher spatial resolution and DL-denoising enhance image quality. Together, these approaches enable efficient, patient-tailored native T1 mapping protocols that expand the clinical applicability of quantitative CMR without compromising measurement accuracy.

Impact: Tracking neurological changes associated with infectious diseases like Nipah virus is challenging because of limited MRI data. We introduce a physics-informed simulation with native noise modelling and a two-stage framework that enhances edge and structural features, improving low-field NiV images.

Impact: RobFuse enhances robustness of SMS reconstruction to calibration mismatch, balancing in-plane artifacts and cross-plane leakage without iterative optimization or deep learning method. It minimizes quantification errors and cost of repeat scans, enabling reliable fMRI analysis for clinicians and researchers.

Impact: The DLR-based T2WI_DL sequence simultaneously achieves ~50% faster acquisition and significantly superior image quality compared to conventional T2WI, effectively resolving the longstanding conflict between speed and quality in clinical prostate MRI.

Impact: The quadratic phase increment and T2-shuffling methods are combined in a portable 110mT system to achieve multi-contrast imaging within an acceptable scan time, enabling single-scan multi-contrast imaging and rapid quantification in low-field MRI.

Impact: Fast, quantitative brain MRI is enabled by reducing phase-cycled bSSFP scan time using compressed sensing and a mode-domain representation of the bSSFP signal. This is paving the way for high-resolution T1 and T2 mapping and broader adoption of bSSFP relaxometry.

Impact: MR elastography examines brain mechanical properties but is inherently long acquisition. We developed a sparsity-based reconstruction enabling accurate high-resolution MRE. We demonstrated the ability to collect whole-brain 1.5mm MRE < 2mins, reducing scan time eightfold while preserving stiffness accuracy < 5% error.

Impact: Ensemble averaging is a straightforward and effective approach to improve the precision and repeatability of 0.55T lung T₁, T₂ and M₀ maps using a Deep Image Prior reconstruction.

Impact: Partial Fourier accelerates low-field MRI while preserving anatomical visibility, as confirmed by radiologists, while undersampling reduced visibility, despite similar SNR/CNR. T1w, T2w, and IR-T1w scans provide complementary information, respectively showing basal ganglia, ventricles/sulci, white matter and cortical regions.

Impact: We proposed a method for increasing temporal resolution of compressed sensing 4D flow, called segment sharing. We demonstrated that it uncovered temporal information of flow measured in a phantom undergoing exercise flow conditions.

Impact: Our findings show that the joint L1+LLR regularization using variable-density Poisson-disc sampling reduces reconstruction error, enhances Z‑spectral consistency, and improves image quality, highlighting its potential for accelerated CEST imaging.

Impact: To enable automated optimization of acquisition parameters via reinforcement learning, we present a GPU-accelerated, EPG-based simulator encapsulated within a Gym-style interactive environment, capable of high-throughput MRF signal trajectory simulation and compatible with reinforcement learning frameworks.

Impact: Physics-informed Deep Learning-based Zero-Shot-models have shown potential in improved fat-water separation with multi-echo MRIs. However, the fat-water swap artifacts in their generated maps cause inaccuracies. This study investigates spatial-smoothness constraints on field-inhomogeneity as a potential idea to cope with swaps.