Density Compensated Unrolled Networks for Non-Cartesian MRI Reconstruction

TL;DR

This paper introduces density compensated unrolled neural networks for non-Cartesian MRI reconstruction, demonstrating improved performance over baselines and providing open-source code for the community.

Contribution

The authors propose a novel neural network architecture that incorporates density compensation for non-Cartesian MRI, addressing a gap in current deep learning approaches.

Findings

Outperforms baseline methods on fastMRI dataset

All components of the proposed design are necessary for optimal performance

Provides open-source implementation of NFFT for TensorFlow

Abstract

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.