Continuous K-space Recovery Network with Image Guidance for Fast MRI Reconstruction

Yucong Meng,Zhiwei Yang,Minghong Duan,Yonghong Shi,Zhijian Song

2024-11-18

Abstract:Magnetic resonance imaging (MRI) is a crucial tool for clinical diagnosis while facing the challenge of long scanning time. To reduce the acquisition time, fast MRI reconstruction aims to restore high-quality images from the undersampled k-space. Existing methods typically train deep learning models to map the undersampled data to artifact-free MRI images. However, these studies often overlook the unique properties of k-space and directly apply general networks designed for image processing to k-space recovery, leaving the precise learning of k-space largely underexplored. In this work, we propose a continuous k-space recovery network from a new perspective of implicit neural representation with image domain guidance, which boosts the performance of MRI reconstruction. Specifically, (1) an implicit neural representation based encoder-decoder structure is customized to continuously query unsampled k-values. (2) an image guidance module is designed to mine the semantic information from the low-quality MRI images to further guide the k-space recovery. (3) a multi-stage training strategy is proposed to recover dense k-space progressively. Extensive experiments conducted on CC359, fastMRI, and IXI datasets demonstrate the effectiveness of our method and its superiority over other competitors.

Image and Video Processing,Artificial Intelligence,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper aims to solve the problem of rapid reconstruction in magnetic resonance imaging (MRI). Specifically, the paper focuses on how to recover high - quality MRI images from undersampled k - space data. Traditional MRI techniques face challenges in clinical applications due to long scanning times. In order to reduce the acquisition time, the goal of fast MRI reconstruction is to recover high - quality images from undersampled k - space data. However, existing methods usually directly apply general deep - learning models to image processing to recover the k - space, ignoring the unique properties of the k - space, resulting in poor recovery effects.

The paper proposes a continuous k - space recovery network (IGKR - Net) based on implicit neural representation (INR) combined with image - domain guidance to improve the performance of MRI reconstruction. The specific contributions are as follows:

1. **Encoder - decoder structure based on INR**: It can directly process k - space signals and reconstruct high - quality k - space.
2. **Image - domain guidance module**: It mines semantic information from low - quality MRI images, guides k - space recovery, and provides rich semantic information.
3. **Multi - stage training strategy**: It gradually reconstructs MRI images from low resolution to high resolution, alleviates the problems of over - smoothing and distortion, and obtains accurate results.

Through extensive experiments on three public datasets, CC359, fastMRI, and IXI, the superiority of this method in fast MRI reconstruction has been proved.