Multi-modal synergistic PET and MR reconstruction using mutually weighted quadratic priors

Area of Science:

• Medical Imaging
• Image Reconstruction
• Multimodal Imaging

Background:

• Positron Emission Tomography (PET) and Magnetic Resonance (MR) imaging are crucial diagnostic tools.
• Noisy PET data and undersampled MR data often lead to suboptimal image quality.
• Improving the quality of simultaneously acquired PET-MR data is essential for accurate diagnosis.

Purpose of the Study:

• To propose a novel framework for synergistic reconstruction of PET-MR and multi-contrast MR data.
• To enhance image quality by addressing noise in PET and undersampling artifacts in MR.
• To develop a method that leverages the strengths of both PET and MR imaging for improved reconstruction.

Main Methods:

• Development of weighted quadratic priors to preserve common boundaries and reduce artifacts.
• Iterative reweighting of priors using normalized multi-modal Gaussian similarity kernels.
• Integration of synergistic reconstruction within PET maximum a posteriori expectation maximization and MR regularized sensitivity encoding algorithms.

Main Results:

• Synergistic reconstructions demonstrated the lowest quantification errors across all modalities compared to conventional methods.
• The proposed method effectively reduced PET Gibbs ringing and MR undersampling artifacts.
• In vivo MR data showed maintained image quality with significant acceleration (3x and 14x), and PET-MR reconstruction showed improved performance over independent methods.

Conclusions:

• The proposed methodology provides a robust framework for multi-modal synergistic image reconstruction.
• This framework can be readily integrated with existing established image reconstruction algorithms.
• The approach offers significant improvements in image quality for combined PET-MR and multi-contrast MR imaging.