Density Compensated Unrolled Networks for Non-Cartesian MRI Reconstruction
This work provides an incremental improvement for MRI reconstruction, specifically for non-Cartesian acquisitions, which is a common problem in medical imaging.
This paper addresses the challenge of reconstructing non-Cartesian MRI acquisitions using deep neural networks. The authors introduce density-compensated unrolled neural networks, which leverage density compensation to correct uneven k-space weighting, demonstrating superior performance compared to baseline methods on the fastMRI dataset.
Deep neural networks have recently been thoroughly investigated as a powerful tool for MRI reconstruction. There is a lack of research, however, regarding their use for a specific setting of MRI, namely non-Cartesian acquisitions. In this work, we introduce a novel kind of deep neural networks to tackle this problem, namely density compensated unrolled neural networks, which rely on Density Compensation to correct the uneven weighting of the k-space. We assess their efficiency on the publicly available fastMRI dataset, and perform a small ablation study. Our results show that the density-compensated unrolled neural networks outperform the different baselines, and that all parts of the design are needed. We also open source our code, in particular a Non-Uniform Fast Fourier transform for TensorFlow.