Memory-efficient Learning for High-Dimensional MRI Reconstruction

TL;DR

This paper introduces a memory-efficient learning framework that reduces GPU memory usage for high-dimensional MRI reconstruction, enabling more complex deep learning models to improve image quality in 3D and 2D+time MRI without significant training time increase.

Contribution

The paper presents a novel memory-efficient learning method that allows training larger deep neural networks for high-dimensional MRI reconstruction with limited GPU memory.

Findings

Improved image reconstruction in 3D MRI and 2D+time cardiac cine MRI.

Significant GPU memory reduction during training.

Marginal increase in training time enabling high-dimensional data processing.

Abstract

Deep learning (DL) based unrolled reconstructions have shown state-of-the-art performance for under-sampled magnetic resonance imaging (MRI). Similar to compressed sensing, DL can leverage high-dimensional data (e.g. 3D, 2D+time, 3D+time) to further improve performance. However, network size and depth are currently limited by the GPU memory required for backpropagation. Here we use a memory-efficient learning (MEL) framework which favorably trades off storage with a manageable increase in computation during training. Using MEL with multi-dimensional data, we demonstrate improved image reconstruction performance for in-vivo 3D MRI and 2D+time cardiac cine MRI. MEL uses far less GPU memory while marginally increasing the training time, which enables new applications of DL to high-dimensional MRI.