Deep Learning-Based Reconstruction for Cardiac MRI: A Review

Abstract

Cardiac magnetic resonance (CMR) is an essential clinical tool for the assessment of cardiovascular disease. Deep learning (DL) has recently revolutionized the field through image reconstruction techniques that allow unprecedented data undersampling rates. These fast acquisitions have the potential to considerably impact the diagnosis and treatment of cardiovascular disease. Herein, we provide a comprehensive review of DL-based reconstruction methods for CMR. We place special emphasis on state-of-the-art unrolled networks, which are heavily based on a conventional image reconstruction framework. We review the main DL-based methods and connect them to the relevant conventional reconstruction theory. Next, we review several methods developed to tackle specific challenges that arise from the characteristics of CMR data. Then, we focus on DL-based methods developed for specific CMR applications, including flow imaging, late gadolinium enhancement, and quantitative tissue characterization. Finally, we discuss the pitfalls and future outlook of DL-based reconstructions in CMR, focusing on the robustness, interpretability, clinical deployment, and potential for new methods.

Artificial Intelligence Applications in Cardiovascular Magnetic Resonance Imaging: Are We on the Path to Avoiding the Administration of Contrast Media?

“…In regression tasks, the algorithm produces a continuous output (such as the ejection fraction values of the left ventricle); in classification tasks, the output is a set of discrete labels (such as the different levels of cardiac contractility). Other tasks include localization tasks, in which the algorithm estimates the coordinates in the image space of bounding boxes around regions of interest (such as anatomical or pathological structures); segmentation tasks, in which the algorithm outputs a class label for each pixel or voxel in the image or volume, typically distinguishing between the background and one or more foreground regions of interest, such as the left ventricle and atrium in non-contrast cine-CMR; acquisition tasks, in which DNNs can be leveraged to adjust imaging parameters in real-time based on patient-specific characteristics, such as heart rate and anatomy, in order to improve spatial resolution, reduce motion artifacts, and enhance the overall image quality; and finally, reconstruction tasks, in which DNNs can be trained to reconstruct high-resolution cardiac images from undersampled k-space data, acquired with accelerated imaging techniques such as compressed sensing [ 28 ].…”

AI in MRI: Computational Frameworks for a Faster, Optimized, and Automated Imaging Workflow

“…More recently, attention has shifted to harnessing DL for accelerating higher-dimensional MRI scans, such as dynamic (temporal) MRI. In this issue, Oscanoa et al provide a comprehensive review of DL-based reconstruction methods for dynamic cardiac MRI, with connections to relevant theory [ 40 ].…”

A motion-corrected deep-learning reconstruction framework for accelerating whole-heart magnetic resonance imaging in patients with congenital heart disease

“…More recently, deep neural networks have been proposed for CMR image reconstruction [21] , [22] , [23] , [24] , [25] , [26] . While compressed sensing and low-rank techniques impose a priori information during the reconstruction, supervised deep-learning-based reconstruction methods instead learn regularizing information from a large set of fully sampled training data.…”