Probabilistic Graphical Models for Dynamic PET: A Novel Approach to Direct Parametric Map Estimation and Image...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Computational Science

Background:

• Conventional dynamic emission tomography uses separate reconstruction and kinetic modeling steps.
• Direct 4D PET reconstruction integrates multiple time frames but often uses deterministic kinetic models.
• Existing methods treat photon counting as the sole source of uncertainty.

Purpose of the Study:

• To develop a probabilistic modeling strategy for dynamic PET reconstruction.
• To incorporate uncertainty in activity time courses into the reconstruction process.
• To create a flexible method that bridges conventional and direct reconstruction approaches.

Main Methods:

• Introduced a hierarchical probabilistic model using graphical modeling.
• Developed an iterative algorithm treating kinetic modeling as prior expectation.
• Enabled control over the trade-off between data-driven and model-driven reconstruction.

Main Results:

• The proposed method generalizes conventional indirect and direct reconstruction approaches.
• Demonstrated improved performance through computer simulations and a real-patient scan.
• Successfully integrated kinetic modeling results as priors for activity time courses.

Conclusions:

• The new probabilistic method offers a flexible and generalized approach to dynamic PET reconstruction.
• It allows for arbitrary kinetic models and priors for parametric maps.
• Provides a unified framework by tuning the influence of kinetic modeling on image reconstruction.