Nonlinear dipole inversion (NDI) enables robust quantitative susceptibility mapping (QSM).

Area of Science:

• Medical Imaging
• Biophysics
• Computational Neuroscience

Background:

• Quantitative Susceptibility Mapping (QSM) is crucial for neuroimaging, but current methods often suffer from over-smoothing or require extensive data.
• Existing reconstruction techniques struggle to balance image quality with data acquisition efficiency.

Purpose of the Study:

• To develop a high-quality Quantitative Susceptibility Mapping (QSM) method using Nonlinear Dipole Inversion (NDI).
• To improve QSM reconstruction by avoiding over-smoothing and enhancing flexibility with limited data.
• To integrate a physics-based model with deep learning for a robust algorithm.

Main Methods:

• Developed Nonlinear Dipole Inversion (NDI) with pre-determined regularization, utilizing a nonlinear forward model with magnitude as a prior.
• Derived a gradient descent update rule for the NDI algorithm.
• Synergistically combined NDI with a Variational Network (VN) to create the VaNDI algorithm, leveraging deep learning without additional parameter tuning.

Main Results:

• NDI achieves image quality comparable to state-of-the-art techniques while preventing over-smoothing.
• NDI demonstrates superior performance to COSMOS, even with a single head orientation (1-direction data).
• Evaluation at 7 T using accelerated Wave-CAIPI acquisitions shows high-quality QSM from as few as 2-direction data at 0.5 mm isotropic resolution.

Conclusions:

• The developed VaNDI algorithm offers high-quality QSM reconstruction with improved efficiency and reduced data requirements.
• NDI provides a flexible and robust approach for QSM, outperforming existing methods.
• This physics-informed deep learning approach advances the field of quantitative MRI.