New preprint on Accelerated MRI Reconstruction

Medical Imaging (MI) often involves the challenging task of reconstructing images from noisy or incomplete measurements, particularly in accelerated Parallel Magnetic Resonance Imaging (MRI). Conventional methods like Compressed Sensing can be time-consuming and yield low-fidelity results. In our latest preprint, we introduce "vSHARP: A DL-Based Approach for Accelerated Parallel MRI Reconstruction." vSHARP leverages the Half-Quadratic Variable Splitting method and employs the Alternating Direction Method of Multipliers (ADMM) to address ill-posed inverse problems in MI. We unroll a differentiable gradient descent process in the image domain and apply a DL-based denoiser, such as a U-Net architecture, to enhance image quality. Additionally, we use a dilated-convolution DL-based model to predict the Lagrange multipliers for the ADMM initialization.

Our experiments demonstrate vSHARP's superiority in accelerated Parallel MRI Reconstruction when compared to state-of-the-art approaches. This work provides insights into the mathematical techniques used and showcases the potential for fast MRI reconstruction. For the full paper, visit here.