WaveSep: A Flexible Wavelet-Based Approach for Source Separation in Susceptibility Imaging.

Area of Science:

• Neuroimaging
• Biophysics
• Medical Physics

Background:

• Separating paramagnetic and diamagnetic susceptibility sources in MRI is crucial for understanding brain function and health.
• Existing deep learning methods for quantitative susceptibility mapping (QSM) have limitations, including requiring high-quality training data and single head orientations.
• No current methods exist for source separation in the more complex susceptibility tensor imaging (STI) framework, which accounts for susceptibility anisotropy.

Purpose of the Study:

• To present a unified and flexible algorithm, WaveSep, for source separation in both QSM and STI.
• To overcome the limitations of existing methods by allowing arbitrary head orientations and not requiring ground-truth training data.
• To enable the estimation of anisotropic second-order susceptibility tensors in STI.

Main Methods:

• WaveSep employs state-of-the-art, data-driven models for dipole inversion using learned proximal operators.
• It then utilizes a Wavelet-based approach for separating paramagnetic and diamagnetic sources without retraining.
• The method accommodates multiple input measurements from arbitrary head orientations for enhanced accuracy.

Main Results:

• WaveSep demonstrates superior performance for susceptibility source separation in QSM compared to existing methods.
• The algorithm achieves unprecedented separation results in the challenging framework of STI.
• Experimental validation was performed on both simulated data and in vivo human brain data.

Conclusions:

• WaveSep offers a flexible and effective solution for paramagnetic and diamagnetic source separation in both QSM and STI.
• The method advances in vivo human brain analysis by providing accurate susceptibility tensor estimation.
• The developed algorithm overcomes key limitations of previous approaches, paving the way for broader applications in neuroimaging research.