Rethinking Dual-Domain Undersampled MRI reconstruction: domain-specific design from the perspective of the receptive field

TL;DR

This paper proposes a novel dual-domain MRI reconstruction model that emphasizes receptive field considerations and introduces domain-specific modules, leading to significant performance improvements over existing methods.

Contribution

It introduces domain-specific modules for dual-domain MRI reconstruction based on receptive field insights, enhancing existing models like DuDoRNet.

Findings

DuDoRNet+ outperforms competing methods on IXI dataset.

Receptive field considerations improve image and K-space interpolation.

Domain-specific modules enhance dual-domain reconstruction quality.

Abstract

Undersampled MRI reconstruction is crucial for accelerating clinical scanning. Dual-domain reconstruction network is performant among SoTA deep learning methods. In this paper, we rethink dual-domain model design from the perspective of the receptive field, which is needed for image recovery and K-space interpolation problems. Further, we introduce domain-specific modules for dual-domain reconstruction, namely k-space global initialization and image-domain parallel local detail enhancement. We evaluate our modules by translating a SoTA method DuDoRNet under different conventions of MRI reconstruction including image-domain, dual-domain, and reference-guided reconstruction on the public IXI dataset. Our model DuDoRNet+ achieves significant improvements over competing deep learning methods.