DIRECT RECONSTRUCTION OF DYNAMIC PET PARAMETRIC IMAGES USING SPARSE SPECTRAL REPRESENTATION

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Dynamic PET requires accurate parametric images for quantitative analysis.
• Conventional methods involve separate image reconstruction and kinetic modeling, which is less efficient.
• Existing direct methods often rely on nonlinear models, necessitating prior knowledge of model order.

Purpose of the Study:

• To develop a direct reconstruction method for dynamic PET that bypasses the need for model order assumption.
• To enhance the statistical efficiency and accuracy of parametric image generation in dynamic PET.

Main Methods:

• A direct reconstruction approach utilizing a linear spectral representation of PET data.
• Incorporation of a Laplacian prior to enforce sparsity in the spectral domain.
• Solving the maximum a posteriori (MAP) formulation using an expectation maximization shrinkage algorithm.
• Development of a bias correction step to refine the MAP estimates.

Main Results:

• The proposed method demonstrates improved bias-variance tradeoff compared to conventional indirect methods.
• Successful generation of parametric images from dynamic PET data without assuming model order.
• Computer simulations validate the enhanced performance of the spectral direct reconstruction approach.

Conclusions:

• The presented linear spectral direct reconstruction method offers a statistically superior alternative for dynamic PET imaging.
• This approach enhances accuracy and efficiency in generating parametric images, crucial for quantitative PET analysis.
• The method's ability to avoid model order assumptions simplifies the process and improves robustness.