A probabilistic Bayesian approach to recover map and phase images for quantitative susceptibility mapping

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Image Reconstruction

Background:

• Undersampling in MRI accelerates data acquisition but often compromises image quality.
• Quantitative Susceptibility Mapping (QSM) requires high-resolution images for accurate results.
• Manual parameter tuning in QSM reconstruction is time-consuming and often impractical in clinical settings.

Purpose of the Study:

• To develop a probabilistic Bayesian approach for reconstructing quantitative susceptibility mapping (QSM) and phase images from undersampled MRI data.
• To enable automatic parameter estimation, thereby avoiding manual tuning and improving image quality.
• To reduce scan time for high-resolution 3D MRI.

Main Methods:

• A novel nonlinear Approximate Message Passing (AMP) framework was developed, incorporating a mono-exponential decay model.
• A sparse prior on wavelet coefficients was interpreted from a Bayesian perspective.
• Parameters were jointly estimated with image wavelet coefficients, and undersampling was performed in the y-z plane of k-space using a Poisson-disk pattern.

Main Results:

• The proposed AMP with Parameter Estimation (AMP-PE) successfully reconstructed $$ maps and phase images for QSM at various undersampling rates.
• AMP-PE demonstrated superior computational efficiency and performance compared to the state-of-the-art L1-norm regularization approach.
• The method achieved comparable or better results than L1 regularization, particularly in scenarios where manual parameter tuning is difficult.

Conclusions:

• AMP-PE leverages both sparse priors and a mono-exponential decay model for enhanced performance in QSM reconstruction.
• The approach eliminates the need for manual parameter tuning, making it suitable for clinical prospective undersampling schemes.
• AMP-PE offers a robust and efficient solution for high-quality QSM from undersampled MRI data.