Multi-contrast reconstruction with Bayesian compressed sensing.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Science

Background:

• Clinical structural MRI often requires multiple image acquisitions with varying contrast preparations.
• Reconstructing high-fidelity images from undersampled k-space data is crucial for efficient MRI.

Purpose of the Study:

• To develop a novel Bayesian compressed sensing algorithm for joint reconstruction of multiple MRI contrasts.
• To improve image reconstruction quality compared to existing methods like M-FOCUSS.

Main Methods:

• A hierarchical Bayesian framework was used for joint inference of image gradient coefficients.
• Hyperparameters representing gradient variance across contrasts were estimated using all images.
• Individual image reconstruction utilized estimated hyperparameters and corresponding k-space data.

Main Results:

• The proposed algorithm achieved higher fidelity reconstructions than individual or M-FOCUSS joint reconstruction.
• Demonstrated up to a four-fold improvement in root-mean-square error.
• Successfully exploited common spatial structures across contrasts without assuming correlation.

Conclusions:

• The hierarchical Bayesian approach enables effective joint reconstruction of multi-contrast MRI data.
• This method offers improved image quality and efficiency for clinical structural MRI.
• The algorithm leverages shared anatomical information across different MRI contrasts.