QSMnet-INR: Single-Orientation Quantitative Susceptibility Mapping via Implicit Neural Representation in k-Space

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Biophysics

Background:

• Quantitative Susceptibility Mapping (QSM) reconstructs magnetic susceptibility from MRI phase data.
• QSM is ill-posed in single-orientation settings due to the dipole kernel's cone-null region.
• Existing methods struggle with stability and artifacts in these challenging scenarios.

Purpose of the Study:

• To develop a physics-informed framework, QSMnet-INR, to address the ill-posed nature of single-orientation QSM.
• To improve the stability and accuracy of QSM reconstructions by integrating implicit neural representations (INRs).
• To reduce artifacts in QSM through enhanced k-space modeling and physical consistency.

Main Methods:

• Proposed QSMnet-INR, integrating an INR into k-space modeling for QSM reconstruction.
• Utilized INR to learn a continuous dipole response, enhancing stability in ill-conditioned regions.
• Implemented a frequency-aware dipole loss to enforce physical model consistency.

Main Results:

• QSMnet-INR demonstrated improved reconstruction quality and reduced artifacts compared to existing methods.
• Significant improvements were observed particularly in single-orientation QSM settings.
• Ablation and sensitivity analyses confirmed the benefits of INR-based modeling and frequency-aware regularization.

Conclusions:

• Integrating implicit representations with physics-informed constraints offers an effective strategy for stabilizing ill-posed QSM reconstruction.
• QSMnet-INR shows promise for enhancing QSM accuracy and reducing artifacts, especially in challenging acquisition settings.
• Further research is needed for extreme susceptibility conditions and novel acquisition parameters.