Data- and Physics-Driven Deep Learning Based Reconstruction for Fast MRI: Fundamentals and Methodologies.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Computational Imaging

Background:

• Magnetic Resonance Imaging (MRI) is crucial for clinical diagnostics.
• Extended MRI scan times negatively impact patient experience and image quality, particularly for volumetric, temporal, and quantitative imaging.
• Accelerating MRI acquisition is a significant challenge in medical imaging.

Purpose of the Study:

• To review recent advancements in MRI acceleration techniques.
• To explore the integration of data-driven and physics-driven models for faster MRI.
• To discuss challenges and future directions in MRI reconstruction.

Main Methods:

• Review of data-driven models including algorithm unrolling, enhancement-based, plug-and-play, and generative models.
• Exploration of physics-informed approaches like parallel imaging and simultaneous multi-slice imaging.
• Analysis of optimized sampling patterns and hardware accelerations.

Main Results:

• Significant progress in MRI acceleration using diverse AI and physics-based models.
• Synergistic integration of data and physics models enhances reconstruction performance.
• Identification of key challenges such as data heterogeneity and model generalization.

Conclusions:

• Advanced MRI acceleration methods show great promise for improving clinical workflows.
• Addressing challenges in data harmonization and federated learning is crucial for broader applicability.
• Future research should focus on enhancing model generalization and performance in real-world MRI scenarios.