Impact: A novel data compression technique is proposed, to significantly reduce MRI data volume in scans accelerated by rapidly modulated B₀ fields (e.g., Wave-CAIPI, FRONSAC, local B₀ coils modulations), enabling compressed-sensing reconstruction for these fast acquisition approaches and facilitating broader adoption.

Impact: MambaRoll enhances image fidelity through improved contextual sensitivity while maintaining high computational efficiency, enabling more reliable reconstructions. This capability can improve the utility of accelerated imaging protocols and learning-based reconstruction in clinical applications.

Impact: SSCU enables high-quality MRI reconstruction without the need for fully sampled reference data. It is particularly effective on low-SNR datasets and has the potential to benefit a wide range of applications, including low-field MRI, rapid clinical scans, and high-resolution imaging.

Impact: By combining physical laws with water content information, the Water Content-Guided Physics Informed Neural Network corrects the overestimation caused by assumptions that are not strictly valid in practice without requiring large amounts of training data, thus improving clinical applicability.

Impact: This work proposes a new reconstruction strategy for scenarios with large phase variations by introducing separate magnitude and phase regularizers in algorithm unrolling with a novel data fidelity approach. Results show reduction of artifacts arising from phase inconsistencies.

Impact: This study demonstrates that Jacobian approximation provides an efficient alternative to standard automatic differentiation for joint optimization of sampling and reconstruction in AutoSamp, enabling faster convergence and comparable reconstruction accuracy for accelerated MRI.

Impact: We propose an uncertainty quantification framework combining quantile regression and conformal prediction with accelerated MRI reconstructions. This yields statistically rigorous uncertainty maps that match the reconstruction error map across acceleration factors, enabling reliable error estimation when the ground-truth is unavailable.

Impact: This work facilitates the adoption of ESPIRiT coil sensitivities into modern MRI reconstructions, especially those that are high-dimensional, high-resolution, and deep learning-based, often run on memory-limited GPUs. Hence, this work can improve the fidelity and robustness of these large-scale reconstructions.

Impact: This work introduces a continuous reconstruction paradigm that decouples image representation from a fixed grid. It avoids pixelation artifacts, enables rendering at arbitrary resolution, and opens new avenues for high-resolution quantitative imaging and understanding fundamental resolution limits.

Impact: We developed a method addressing the primary limitation of combined diffusion-relaxometry MRI – long scan time - by enabling high acceleration factor (12x). This advancement facilitates the broader adoption of advanced microstructural imaging techniques in basic and clinical neuroscience research.