[Comparison of the number of parameters using nonlinear iteration methods for compartment model analysis with 18F-FDG...

In the analysis of the brain kinetic images with 18F-FDG, compartment models with 3 or 4 parameters and an estimation method of nonlinear least squares (NLS) have been applied. We compared the Simplex method and the modified Marquardt method as an iterative NLS algorithm on the simulation data with variable noise. The estimated k-values were equal in the both algorithms. The calculation with modified Marquardt method was three times faster than that of the Simplex method. The dependence of modified Marquardt estimates on the initial values was larger than that of the Simplex, but the permissible initial value range was wide enough for clinical data. Therefore, the modified Marquardt method was found to be more suitable for the compartment model analysis of 18F-FDG than the Simplex method. We also compared the 3-parameter with 4-parameter models for 45 min and 120 min data acquired after administration. The 3-parameter model was suitable for the 45-minute data acquisition, and the 4-parameter model was suitable for the 120-minute data acquisition.