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

The Akaike information criterion and the associated Akaike weights (AW) rank pharmacokinetic models on the basis of goodness-of-fit and number of parameters. The usefulness of this information for improving the haemodynamic parameter estimates from DCE-MRI was investigated through two examples. In each of these, the estimates from the two-compartment exchange model (2CXM) were combined on the basis of the AW with those of a simplified model (either the uptake model or the extended Tofts model). Data were simulated using the 2CXM for a range of experimental and tissue conditions. Two multimodel approaches exploiting the AW were investigated: the ‘bestmodel’ approach which selects the parameter estimates from the model with highest AW and the ‘weighted model’ approach in which AW-weighted averages of the estimates from the competing models are calculated. Although these approaches were shown to be beneficial in some cases, they were found to frequently lead to unexpected increases in the bias and/or uncertainty of the resulting parameter estimates. Within the limited scope of this simulation study, the use of the Akaike criterion showed no systematic benefit over a fitting strategy involving only the more complex model.