Self-Supervised Physics-Guided Deep Learning Reconstruction For High-Resolution 3D LGE CMR

TL;DR

This paper introduces a self-supervised physics-guided deep learning method for high-resolution 3D LGE cardiac MRI, enabling faster scans without fully-sampled data and outperforming traditional compressed sensing at higher acceleration factors.

Contribution

It extends self-supervised physics-guided deep learning to 3D LGE CMR, addressing small dataset challenges and improving acceleration performance.

Findings

Proposed method achieves 6-fold acceleration in 3D LGE CMR.

Outperforms traditional compressed sensing at 3-fold acceleration.

Validated through results and reader study.

Abstract

Late gadolinium enhancement (LGE) cardiac MRI (CMR) is the clinical standard for diagnosis of myocardial scar. 3D isotropic LGE CMR provides improved coverage and resolution compared to 2D imaging. However, image acceleration is required due to long scan times and contrast washout. Physics-guided deep learning (PG-DL) approaches have recently emerged as an improved accelerated MRI strategy. Training of PG-DL methods is typically performed in supervised manner requiring fully-sampled data as reference, which is challenging in 3D LGE CMR. Recently, a self-supervised learning approach was proposed to enable training PG-DL techniques without fully-sampled data. In this work, we extend this self-supervised learning approach to 3D imaging, while tackling challenges related to small training database sizes of 3D volumes. Results and a reader study on prospectively accelerated 3D LGE show that the proposed approach at 6-fold acceleration outperforms the clinically utilized compressed sensing approach at 3-fold acceleration.