Emerging Trends in Fast MRI Using Deep-Learning Reconstruction on Undersampled k-Space Data: A Systematic Review

Abstract

Magnetic Resonance Imaging (MRI) is an essential medical imaging modality that provides excellent soft-tissue contrast and high-resolution images of the human body, allowing us to understand detailed information on morphology, structural integrity, and physiologic processes. However, MRI exams usually require lengthy acquisition times. Methods such as parallel MRI and Compressive Sensing (CS) have significantly reduced the MRI acquisition time by acquiring less data through undersampling k-space. The state-of-the-art of fast MRI has recently been redefined by integrating Deep Learning (DL) models with these undersampled approaches. This Systematic Literature Review (SLR) comprehensively analyzes deep MRI reconstruction models, emphasizing the key elements of recently proposed methods and highlighting their strengths and weaknesses. This SLR involves searching and selecting relevant studies from various databases, including Web of Science and Scopus, followed by a rigorous screening and data extraction process using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. It focuses on various techniques, such as residual learning, image representation using encoders and decoders, data-consistency layers, unrolled networks, learned activations, attention modules, plug-and-play priors, diffusion models, and Bayesian methods. This SLR also discusses the use of loss functions and training with adversarial networks to enhance deep MRI reconstruction methods. Moreover, we explore various MRI reconstruction applications, including non-Cartesian reconstruction, super-resolution, dynamic MRI, joint learning of reconstruction with coil sensitivity and sampling, quantitative mapping, and MR fingerprinting. This paper also addresses research questions, provides insights for future directions, and emphasizes robust generalization and artifact handling. Therefore, this SLR serves as a valuable resource for advancing fast MRI, guiding research and development efforts of MRI reconstruction for better image quality and faster data acquisition.

Keywords: Deep Bayesian Learning; Magnetic Resonance Imaging; acquisition time reduction; compressive sensing; deep MRI reconstruction; deep dictionary learning; deep learning; fast MRI; k-space; parallel MRI.

Figure 7

Self-Supervised Training Paradigm for Unrolled MRI Reconstruction Network: Regularizer (R) and Data Consistency (DC) Components (adapted with changes from [137]).

Figure 1

Fast MRI Knee Image Reconstruction: From Fully Sampled to Aliased Images: (a) Reconstructed MRI obtained from DL method (X), (b) $y^{\mathrm{Full}}$, (c) $X^{\mathrm{Full}}$, (d) M, (e) y, and (f) $X^{\mathrm{aliased}}$ (adapted with changes from [5]).

Figure 2

Methodology and Criteria for Inclusion and Exclusion of Research Articles from WoS and Scopus Databases: A PRISMA Guideline-Based Approach, Augmented with a Manual Search on Google Scholar.

Figure 3

Residual Learning-based MRI Reconstruction Process (adapted with changes from [61]).

Figure 4

Encoder-Decoder-based MRI Reconstruction Method (adapted with changes from [62]).

Figure 5

Unrolled networks: Mapping an Iterative Algorithm into a (a) Deep Network with (b) Trainable Parameters (in Blue) (adapted with changes from [75]).

Figure 6

Generative Adversarial Network (GAN)-based MRI Reconstruction Process.

Figure 8

Diagrammatic Flow of SuperMAP-based End-to-End Quantitative Mapping (adapted with changes from [39]).

Figure 9

Deep Learning Model for Tissue Quantification in MRF with Spatially Constrained Quantification (SCQ) Methodology (adapted with changes from [182]).

Figure 10

Publication Analysis of DL-based MRI Reconstruction including Compressive Sensing (CS) only, CS combined with Deep Learning (DL), and DL-only (2018–2023).

Figure 11

Number of Published Papers on DL Reconstruction Architectures focusing on Residual Learning (RL), Image Representation using Encoders and Decoders (IR-ED), Data-consistency Layers and Unrolled Networks (DC-UN), Learned Activations and Attention Modules (LA-AM), and Plug-and-play Priors, Diffusion Models, and Bayesian Methods (PDBM).

Figure 12

Trends on Improving Reconstruction-Related MRI Applications including Non-Cartesian Reconstruction (NCR), Super-resolution (SR), Joint learning: Coil-sensitivity and Reconstruction (JL-Coil), Joint learning: Sampling and Reconstruction (JL-Samp), Quantitative Mapping (QM), MR Fingerprinting (MRF), and Dynamic MRI (DMRI).