DCE-MRI protocol for constraining absolute pharmacokinetic modeling errors within specific accuracy limits.

PURPOSE

To quantify the effects of DCE acquisition and pharmacokinetic modeling processing methodologies on the absolute accuracy and precision of derived pharmacokinetic (PK) parameter values using a novel anthropomorphic phantom test device in which "ground truth" values were known a priori.

METHODS

Ground truth arterial input function (AIF), tumor, and healthy tissue contrast agent concentration-time curves (CTCs) were established within the phantom and repeatedly measured on a 3T MRI scanner with varying temporal resolution (T  = 1.22-30.6 s). Ground truth CTCs, K , v , and k values were directly compared to measured values as a function of T , with and without the application of voxel-wise flip-angle corrections applied to the data and PK modeling performed using linear and nonlinear forms of the standard Tofts model.

RESULTS

Measurement of the AIF was strongly affected by the T used (AIF curve-shape feature errors: 3%-222% for T : 1.22-30.6 s), which directly translated to errors in the derived K , v , and k values of 1%-24%, 2%-5%, and 1%-26% respectively across this T range (flip-angle correction applied). Further appreciable improvements in accuracy and precision arising from the use of flip angle corrections and nonlinear least squares fitting were quantified and used to identify optimal acquisition and analysis methodologies for which measurement errors could be constrained below threshold levels.

CONCLUSION

This quantitative study provides insight into how errors in AIF measurement propagate to errors in PK parameter outputs. Absolute quantification of the accuracy and precision of MR-measured CTCs, and resultant PK parameter values, allowed for an optimal temporal resolution to be defined commensurate with maintaining K , v , and k measurement errors below 5% and 10% levels. An appreciable gain in PK parameter estimation accuracy at the analysis stage was also demonstrated using flip-angle corrections and a linear approach to PK model fitting.