Joint diffusion: mutual consistency-driven diffusion model for PET-MRI co-reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Positron Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) systems provide complementary functional and anatomical data.
• PET imaging faces challenges with low signal-to-noise ratio, while MRI acquisition is time-consuming.
• Reducing MRI data collection (k-space) to save time often compromises image quality.

Purpose of the Study:

• To enhance the image quality of combined PET-MRI scans.
• To address the trade-off between acquisition speed and image quality in PET-MRI.
• To leverage the inherent complementarity of PET and MRI data for improved reconstruction.

Main Methods:

• A novel Bayesian framework PET-MRI joint reconstruction model, MC-Diffusion, was developed.
• The model transforms joint reconstruction into a joint regularization problem with independent data fidelity terms.
• A joint score-based diffusion model was employed to learn the joint probability distribution of PET and MRI data.

Main Results:

• The MC-Diffusion model demonstrated qualitative and quantitative improvements in PET-MRI image reconstruction.
• Comparative analysis on the ADNI dataset showed superior performance against LPLS and Joint ISAT-net.
• The model effectively enhanced the quality of both PET and MRI components of the scans.

Conclusions:

• The MC-Diffusion model successfully enhances PET-MRI image quality by integrating modality principles and exploiting data complementarity.
• Utilizing diffusion models to learn joint probability distributions elucidates latent correlations between PET and MRI.
• This approach offers a deeper understanding of deep learning priors compared to black-box methods.