Unmixing dynamic PET images with variable specific binding kinetics.

Area of Science:

• Nuclear Medicine
• Medical Imaging Analysis
• Biophysics

Background:

• Dynamic positron emission tomography (PET) imaging is crucial for understanding tracer kinetics in tissues.
• Conventional multivariate analysis methods like PCA, ICA, and NMF often fail to capture nonlinear variations in time-activity curves.
• This limitation hinders the reliable and interpretable recovery of kinetic behavior, especially in tissues with specific binding.

Purpose of the Study:

• To develop an alternative analysis paradigm for dynamic PET images that addresses the limitations of conventional methods.
• To explicitly model spatial fluctuations in tracer exchange rates between free and specifically bound compartments.
• To improve the reliability, understandability, and interpretability of dynamic PET data analysis.

Main Methods:

• Proposed an analysis paradigm based on linear unmixing, adapted from hyperspectral imaging.
• Combined nonnegative matrix factorization (NMF) with a sum-to-one constraint for an exhaustive data description.
• Explicitly modeled spatial variability of specific binding tissue signatures using a perturbed component.

Main Results:

• The proposed method successfully accounts for spatial fluctuations in tracer exchange rates.
• Performance was validated on both synthetic and real dynamic PET imaging data.
• The new approach demonstrated favorable performance compared to conventional analysis methods.

Conclusions:

• The novel linear unmixing-based method offers a more robust and interpretable approach to dynamic PET image analysis.
• It effectively captures nonlinear kinetic behaviors and spatial variability, overcoming limitations of traditional techniques.
• This method holds promise for advancing the quantitative analysis of PET imaging studies.