Comparison of different population-averaged arterial-input-functions in dynamic contrast-enhanced MRI of the prostate: Effects on pharmacokinetic parameters and their diagnostic performance

Abstract Purpose To assess the effect of different population-averaged arterial-input-functions (pAIF) on pharmacokinetic parameters from dynamic contrast-enhanced MRI (DCE-MRI) and their diagnostic accuracy regarding the detection of potentially malignant prostate lesions. Materials and methods 66 male patients (age 65.4 ± 10.8y) with suspected prostate cancer underwent multiparametric MRI of the prostate including T2-w, DWI-w and DCE-MRI sequences at a 3 T MRI scanner. All detected lesions were categorized based on ACR PI-RADS version 2 and divided into 2 groups (A: PI-RADS ≤ 3, n = 32; B: PI-RADS > 3, n = 34). In each DCE-MRI dataset, pharmacokinetic parameters (Ktrans, Kep and ve) and goodness of fit (chi2 ) were generated using the Tofts model with 3 different pAIFs (fast, intermediate, slow) as provided by a commercially available postprocessing software. Pharmacokinetic parameters, their diagnostic accuracies and model fits were compared for the 3 pAIFs. Results Ktrans, Kep and ve differed significantly among the 3 pAIFs (all p < .001). Ktrans and Kep were significantly higher in group B compared to group A (all p < .001). For chi2 , lowest results (representing highest goodness of fit) were found for intermediate pAIF (chi2 0.073). ROC analyses revealed comparable diagnostic accuracies for the different pAIFs, which were high for Ktrans and Kep and low for ve. Conclusion Choosing various pAIF types causes a high variability in pharmacokinetic parameter estimates. Therefore, it is of great importance to consider this as potential artifact and thus keep AIF type selection constant in DCE-MRI studies.