DF-QSM: Data Fidelity based Hybrid Approach for Improved Quantitative Susceptibility Mapping of the Brain

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Biophysics

Background:

• Quantitative Susceptibility Mapping (QSM) is a crucial MRI technique for tissue magnetic susceptibility quantification.
• Deep learning (DL) shows promise in QSM but often suffers from limited generalizability due to data bias.
• Existing DL-based QSM methods require refinement for broader applicability.

Purpose of the Study:

• To develop a hybrid, two-step reconstruction method to improve the generalizability of DL-based QSM.
• To refine DL-predicted susceptibility maps for enhanced consistency with measured MR fields.
• To validate the proposed method against existing QSM techniques.

Main Methods:

• A hybrid two-step reconstruction framework integrating DL predictions with local field consistency.
• Refinement of DL-generated susceptibility maps within the developed framework.
• Validation using brain MRI datasets and comparison with established DL and model-based DL QSM methods.

Main Results:

• The hybrid method demonstrated improved QSM reconstruction quality across various MRI acquisition settings.
• Enhanced performance was observed even with DL models trained on limited data.
• The approach successfully refined susceptibility map predictions for better field consistency.

Conclusions:

• The proposed hybrid reconstruction approach significantly enhances the generalizability and accuracy of DL-based QSM.
• This method offers a robust solution for improving QSM in diverse clinical and research scenarios.
• The findings suggest a promising direction for more reliable quantitative MRI.