Factor Analysis of Dynamic PET Images: Beyond Gaussian Noise.

Area of Science:

• Medical Imaging
• Quantitative Imaging
• Data Analysis

Background:

• Dynamic Positron Emission Tomography (PET) imaging generates complex data.
• Extracting tissue time-activity curves (TACs) is crucial for quantitative analysis.
• Unsupervised factor analysis is a relevant tool for TAC extraction, but sensitive to noise.

Purpose of the Study:

• To investigate the utility of various divergence measures within a factor analysis framework for dynamic PET.
• To evaluate the impact of different β-divergence values on the interpretability and reliability of TACs.
• To assess the performance across various noise distributions, including Gaussian, Poissonian, and Gamma.

Main Methods:

• Employed factor analysis with a focus on data-fitting terms designed using β-divergence.
• Tested three distinct factor models incorporating β-divergence.
• Evaluated algorithm performance using synthetic and real dynamic PET image data.
• Examined a range of β values to cover different noise characteristics.

Main Results:

• The study demonstrates that β-divergence can be effectively integrated into factor analysis for dynamic PET.
• Non-standard values of β showed improved performance in extracting TACs compared to standard approaches.
• The choice of β significantly influences the analysis, particularly in the presence of complex noise.

Conclusions:

• β-divergence offers a flexible approach to modeling noise in dynamic PET factor analysis.
• Optimizing β values can enhance the accuracy and reliability of extracted tissue time-activity curves.
• This method provides a promising avenue for improved unsupervised analysis of dynamic PET data.