Linear least squares compartmental-model-independent parameter identification in PET

Area of Science:

• Nuclear Medicine
• Biophysics
• Computational Biology

Background:

• Dynamic scan data analysis in nuclear imaging often relies on complex compartmental models.
• Traditional iterative nonlinear least squares methods can be prone to convergence failures and false solutions.
• Simplified analytical approaches are needed for efficient and robust parameter estimation.

Purpose of the Study:

• To develop and validate a simplified linear regression approach for analyzing dynamic scan data.
• To compare the performance of linear regression fitting with traditional iterative nonlinear least squares methods.
• To evaluate the impact of data noise on parameter estimation using Monte Carlo simulations.

Main Methods:

• A linear-regression straight-line parameter fitting method was developed for specific and nonspecific models.
• Multiple linear regression was employed using spreadsheet software for data fitting.
• Positron emission tomography (PET)-acquired gray-matter images from dynamic scans were analyzed.
• Simulated and patient data were used for validation, alongside Monte Carlo simulations for error analysis.

Main Results:

• The linear regression method demonstrated agreement with traditional iterative nonlinear least squares.
• Monte Carlo simulations indicated small noise-induced biases and quantifiable parameter standard deviations.
• Unique straight-line graphical displays facilitated visualization of data influences on macroparameters.

Conclusions:

• Linear regression offers a simple, fast, and easily implementable alternative for dynamic scan data analysis.
• This method avoids convergence issues inherent in iterative least squares, providing reliable results.
• The approach enables robust multiparameter, model-independent analyses, enhancing understanding of biological systems.