[Research progress on quantitative magnetic susceptibility imaging reconstruction method based on improved U-network...

Area of Science:

• Medical Imaging
• Computational Biology
• Biophysics

Background:

• Quantitative magnetic susceptibility imaging (QSM) reconstructs tissue magnetic susceptibility from MRI phase data.
• The dipole inversion stage is critical but prone to artifacts and deviations with traditional methods.
• Deep learning, particularly U-Net architectures, offers potential to overcome these limitations.

Purpose of the Study:

• To summarize recent advancements (2020-present) in U-Net based models for QSM dipole inversion.
• To categorize and analyze different U-Net architectural improvements for QSM.
• To forecast future trends in deep learning for QSM.

Main Methods:

• Review and categorization of U-Net based models applied to QSM dipole inversion.
• Analysis of structural optimization, physical constraints, and generalization ability improvements.
• Synthesis of current research to identify developmental trajectories.

Main Results:

• U-Net models significantly improve dipole inversion by mitigating artifacts and deviations.
• Categorization reveals diverse strategies: structural optimization, physical constraint integration, and generalization enhancement.
• Identified trends point towards more robust and accurate QSM reconstruction.

Conclusions:

• Improved U-Net models are crucial for overcoming dipole inversion challenges in QSM.
• Enhanced QSM accuracy through deep learning supports improved medical image analysis.
• Future developments are expected to further refine QSM for clinical applications and disease diagnosis.