Magnetic Resonance Fingerprinting: The Role of Artificial Intelligence

Abstract

Cardiovascular magnetic resonance (CMR) imaging is a powerful tool for assessing the function and structure of the heart. An emerging application of CMR is quantitative tissue characterization of the myocardial substrate, which can potentially provide earlier and more sensitive detection of various pathologies than conventional qualitative imaging. Some of the most commonly measured tissue properties are the MRI relaxation time constants T1 and T2. Recently, novel methods including cardiac Magnetic Resonance Fingerprinting (MRF) have been proposed to simultaneously quantify multiple tissue properties during a single rapid acquisition. By combining a fast undersampled data acquisition with dictionary-based pattern matching, cardiac MRF has the potential to streamline CMR exams and provide highly accurate, precise, and reproducible measurements. However, the processes of MRF dictionary generation and pattern matching can be time-consuming and memory-intensive, especially in applications that seek to quantify a large number of tissue properties simultaneously or that require frequent dictionary generation. The combination of deep learning methods with MRF is an emerging research field that may address many of the limitations of dictionary-based MRF and may facilitate the clinical translation of novel cardiac MRF technology. This chapter will begin by providing an overview of conventional methods for CMR tissue parameter mapping before introducing the concept of MRF. We will discuss some challenges associated with current implementations of cardiac MRF and how these may be overcome using artificial intelligence (AI), including a review of several state-of-the-art deep learning methods.KeywordsDeep learningMR fingerprinting (MRF)Neural networksNon-CartesianRelaxometryTissue characterization