Model selection criteria for dynamic brain PET studies.

Area of Science:

• Nuclear medicine
• Radiochemistry
• Pharmacokinetics

Background:

• Optimal kinetic modeling is crucial for quantitative analysis in brain PET studies.
• Several criteria exist to select the best kinetic model, but their agreement is not fully established.
• This study compares five common model selection criteria: AIC, AIC_C, MSC, SC, and F-test.

Purpose of the Study:

• To evaluate the consistency of kinetic model preference identification using five different model selection criteria.
• To determine if discrepancies among criteria affect the final model choice in brain PET studies.

Main Methods:

• Six tracers ([¹¹C]FMZ, [¹¹C]GMOM, [¹¹C]PK11195, [¹¹C]Raclopride, [¹⁸F]FDG, [¹¹C]PHT) were analyzed.
• Time-activity curves (TACs) were modeled using six plasma input models (1T2k, 2T3k, 2T4k, with and without V_B).
• Model preference percentages were calculated for each tracer and criterion, with F-test as the reference.

Main Results:

• All five model selection criteria demonstrated strong agreement in identifying preferred kinetic models across tracers.
• The F-test confirmed AIC-preferred models in 87% of cases.
• Minimal disagreement was observed between AIC and AIC_C, with consistent model selection across all criteria except for one subject.

Conclusions:

• The evaluated model selection criteria (AIC, AIC_C, MSC, SC, F-test) yield largely consistent results for brain PET kinetic modeling.
• Minor differences among criteria do not significantly alter the overall selection of the optimal kinetic model.
• This suggests robustness in quantitative analysis regardless of the specific criterion used.