PET image reconstruction using multi-parametric anato-functional priors.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Conventional anatomical priors in brain PET reconstruction are limited by PET-MR mismatches.
• Existing methods struggle with partial volume correction and low-count PET data.
• Need for advanced priors that integrate multi-parametric information for improved accuracy.

Purpose of the Study:

• To investigate multi-parametric anato-functional (MR-PET) priors for maximum a posteriori (MAP) reconstruction of brain PET data.
• To address limitations of conventional priors in the presence of PET-MR mismatches.
• To evaluate the suitability of these priors for low-count PET data reconstruction.

Main Methods:

• Unified framework for conventional (Tikhonov, TV) and state-of-the-art anatomical priors (Kaipio, Bowsher, Gaussian).
• Extension of Bowsher and Gaussian priors to multi-parametric versions to handle PET-MR mismatches.
• Proposal of a modified joint Burg entropy prior exploiting all parametric information.
• Evaluation using 3D simulations with introduced mismatches and clinical [18F]florbetapentabene and [18F]FDG PET datasets.

Main Results:

• Joint Burg entropy prior outperformed conventional priors in preserving PET-unique lesions.
• Multi-parametric Gaussian and Bowsher priors enhanced edge preservation and bias-variance performance.
• MAP reconstruction of low-count PET data with joint entropy prior achieved comparable quality to ML reconstruction with 5x more counts.
• TV and proposed priors preserved simulated tumors in FDG dataset.

Conclusions:

• Multi-parametric anato-functional priors effectively address pitfalls of conventional priors in MR-guided PET reconstruction.
• These priors enhance diagnostic confidence by improving image quality and lesion detection.
• The joint Burg entropy prior shows significant promise for low-count PET data.