Artificial intelligence in cardiac magnetic resonance fingerprinting

Abstract

Magnetic resonance fingerprinting (MRF) is a fast MRI-based technique that allows for multiparametric quantitative characterization of the tissues of interest in a single acquisition. In particular, it has gained attention in the field of cardiac imaging due to its ability to provide simultaneous and co-registered myocardial T₁ and T₂ mapping in a single breath-held cardiac MRF scan, in addition to other parameters. Initial results in small healthy subject groups and clinical studies have demonstrated the feasibility and potential of MRF imaging. Ongoing research is being conducted to improve the accuracy, efficiency, and robustness of cardiac MRF. However, these improvements usually increase the complexity of image reconstruction and dictionary generation and introduce the need for sequence optimization. Each of these steps increase the computational demand and processing time of MRF. The latest advances in artificial intelligence (AI), including progress in deep learning and the development of neural networks for MRI, now present an opportunity to efficiently address these issues. Artificial intelligence can be used to optimize candidate sequences and reduce the memory demand and computational time required for reconstruction and post-processing. Recently, proposed machine learning-based approaches have been shown to reduce dictionary generation and reconstruction times by several orders of magnitude. Such applications of AI should help to remove these bottlenecks and speed up cardiac MRF, improving its practical utility and allowing for its potential inclusion in clinical routine. This review aims to summarize the latest developments in artificial intelligence applied to cardiac MRF. Particularly, we focus on the application of machine learning at different steps of the MRF process, such as sequence optimization, dictionary generation and image reconstruction.

Keywords: artificial intelligence (AI); cardiac MRF; cardiac magnetic resonance (CMR); magnetic resonance fingerprinting (MRF); multiparametric imaging.

Figure 1

An overview of cardiac MRF framework. (A1) Repetition time (TR) and variable flip angles (FA) may be pseudo-randomly varied throughout acquisition. (A2) Magnetization preparation pulses are introduced to increment contrast weighing on the desired parameters (in this example Inversion Recovery, (IR pulses, in red), T₂ preparation, (T₂ prep pulses in green) and T_1ρ preparation, (T_1ρ prep pulses in blue) are included to encode T₁ and T₂ contrasts before some heartbeats). (B) Highly undersampled images are obtained, and (C) a subject-specific dictionary due to the unique cardiac rhythm during the scan is calculated in parallel. (D) Matching the temporal evolution of the signal measured with the dictionary will provide (E) inherently co-registered parametric maps of the scanned region. The different colored dots in (A1) correspond to different timepoints and contrasts (B) during the sequence.

Figure 2

Left: Artificial Intelligence (AI) encompasses tasks performed by machines and computers that would normally require human intelligence. A subfield of AI is Machine Learning (ML), a technique whereby computer algorithms learn to perform a task from training data rather than requiring explicitly pre-programmed rules, allowing them to provide predictions for unseen examples. Deep Learning (DL) is a subfield of ML that uses artificial Neural Networks (NNs), modeled on neurons in the human brain, trained using data to provide predictions. Right: NN architectures used in cardiac MRF. Feedforward neural networks consist of an input later which could be for example a fingerprint in MRF, followed by a series of hidden layers, followed by a final output later that could output for example tissue parameters. RNNs are ideal for sequence data and take an input timepoint, along with an internal hidden state that encodes information from previous data in the sequence, thus incorporating memory of previous patterns in the sequence.

Figure 3

The cardiac MRF workflow from sequence design and optimization through to parameter map estimation and potential uses of cardiac MRF data, where DL-based analyses such as Radiomics can be used to provide a diagnosis or predict outcomes. Black arrows indicate the flow of steps taken in the cardiac MRF workflow. The red dashed arrows indicate steps where DL methods have or could be applied within this workflow.

Figure 4

Parameter maps generated by a NN for cardiac MRF from Hamilton et al. (82) for two healthy subjects and compared to maps generated using dictionary matching. The feedforward network with skip connections that was used takes the real and imaginary components of the fingerprint and the RR intervals as input and outputs parameter estimates for T₁ and T₂ on a fingerprint-wise basis. The network produces accurate parameter estimations for different cardiac rhythms, even for subject B, with a variable heart rate and 1 missed ECG-trigger.

Figure 5

Spatially-regularized convolutional neural network from Balsiger et al. (90). In addition to the temporal information from each fingerprint, for each voxel a HxWxT patch of undersampled image data is used to calculate the parameter maps [in their work, Balsiger et al. (90) employ a patch size of H = W = 15]. The network achieves improved accuracy by incorporating spatial regularization in the map generation process from undersampled image data.