Unsupervised MRI Reconstruction with Generative Adversarial Networks

TL;DR

This paper introduces an unsupervised deep learning framework using generative adversarial networks for MRI reconstruction, eliminating the need for fully-sampled training data and demonstrating improved anatomical recovery in undersampled MRI scans.

Contribution

It presents a novel unsupervised GAN-based approach for MRI reconstruction that works without fully-sampled ground truth data, applicable to various undersampled MRI scenarios.

Findings

Outperforms conventional methods in anatomical detail recovery

Effective in both retrospective and prospective undersampling scenarios

Reduces reliance on fully-sampled training data

Abstract

Deep learning-based image reconstruction methods have achieved promising results across multiple MRI applications. However, most approaches require large-scale fully-sampled ground truth data for supervised training. Acquiring fully-sampled data is often either difficult or impossible, particularly for dynamic contrast enhancement (DCE), 3D cardiac cine, and 4D flow. We present a deep learning framework for MRI reconstruction without any fully-sampled data using generative adversarial networks. We test the proposed method in two scenarios: retrospectively undersampled fast spin echo knee exams and prospectively undersampled abdominal DCE. The method recovers more anatomical structure compared to conventional methods.