The Akaike information criterion in DCE-MRI: does it improve the haemodynamic parameter estimates?

Area of Science:

• Pharmacokinetics and pharmacokinetic modeling
• Medical imaging analysis
• Biophysical modeling

Background:

• Akaike information criterion (AIC) and Akaike weights (AW) are used to rank pharmacokinetic models based on fit and complexity.
• Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is crucial for assessing tissue hemodynamics.
• Combining estimates from multiple pharmacokinetic models using AW is a potential strategy to improve parameter accuracy.

Purpose of the Study:

• To investigate the utility of Akaike information criterion (AIC) and Akaike weights (AW) for enhancing hemodynamic parameter estimation from DCE-MRI.
• To evaluate two multimodel approaches: 'bestmodel' selection and 'weighted model' averaging.

Main Methods:

• Pharmacokinetic models, including the two-compartment exchange model (2CXM), uptake model, and extended Tofts model, were utilized.
• Data were simulated using the 2CXM under various experimental and tissue conditions.
• Two multimodel approaches combining model estimates based on AW were explored: 'bestmodel' and 'weighted model'.

Main Results:

• Both 'bestmodel' and 'weighted model' approaches, while beneficial in some instances, frequently led to increased bias and/or uncertainty in parameter estimates.
• Simulation results indicated that the Akaike criterion did not offer a systematic advantage over using only the more complex two-compartment exchange model (2CXM).

Conclusions:

• The application of Akaike information criterion (AIC) and Akaike weights (AW) in multimodel approaches for DCE-MRI parameter estimation may not consistently improve results.
• The findings suggest that relying solely on a more complex model, like the 2CXM, might be preferable to multimodel averaging or selection strategies based on AIC in certain scenarios.