Enhancing uncertainty assessment in dynamic PET imaging with residual permutation and clustering

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Computational Science

Background:

• Quantitative Positron Emission Tomography (PET) is vital for disease diagnosis and therapy monitoring.
• Reliable kinetic parameter estimation in PET requires robust uncertainty quantification.
• Current methods like Bayesian, bootstrap, and deep learning have limitations in computational cost, noise sensitivity, or data requirements.

Purpose of the Study:

• To develop a computationally efficient and robust framework for uncertainty estimation in dynamic PET.
• To address the limitations of existing methods for quantifying uncertainty in kinetic parameters.
• To provide a clinically deployable solution for uncertainty-aware dynamic PET analysis.

Main Methods:

• Proposed a clustering-based Residual Permutation (RP) framework for uncertainty estimation in dynamic PET.
• Generated pseudo time-activity curves (TACs) by permuting fitting residuals within kinetically homogeneous clusters.
• Introduced a regularized model with Huber loss and elastic-net regularization for improved numerical stability and overfitting prevention.

Main Results:

• The RP framework provides uncertainty estimates that scale consistently with noise levels.
• Preserved expected physical disparities between kinetic parameters across heterogeneous regions.
• Demonstrated robust performance on simulated data (TACs and XCAT-OSEM reconstructions) and clinical total-body PET data.

Conclusions:

• The RP framework offers a distribution-free, training-independent, and computationally efficient solution for voxel-wise uncertainty quantification in dynamic PET.
• It bridges the gap between computationally intensive Bayesian methods and data-hungry deep learning approaches.
• RP is a robust and clinically deployable method for uncertainty-aware dynamic PET analysis.