Machine learning-based kinetic modeling: a robust and reproducible solution for quantitative analysis of dynamic PET...

Area of Science:

• Nuclear Medicine
• Medical Imaging Analysis
• Computational Biology

Background:

• Dynamic Positron Emission Tomography (PET) data analysis commonly employs compartment models.
• Traditional iterative fitting (IF) methods can suffer from parameter overfitting and poor reproducibility, particularly with noisy data.

Purpose of the Study:

• To introduce a novel machine learning (ML) based kinetic modeling method for quantitative analysis of dynamic PET data.
• To address the limitations of traditional IF methods, enhancing robustness and reproducibility.

Main Methods:

• Developed an ML-based kinetic modeling approach utilizing a historical reference database.
• Implemented a multi-thread grid parameter searching technique for automatic model adjustment.
• Proposed a candidate competition concept to integrate ML and IF methods for balanced data fitting.

Main Results:

• The ML method directly handles noisy PET data without image smoothing.
• Achieved robust and reproducible quantitative analysis for both VOI-based and pixel-wise assessments.
• The candidate competition concept balances fitting to historical and unseen target curves.

Conclusions:

• The proposed ML-based kinetic modeling offers a user-independent and superior alternative to traditional methods for dynamic PET data analysis.
• This approach enhances the reliability and accuracy of quantitative PET imaging.
• The method shows significant potential for advancing precision in nuclear medicine applications.