Repeatability of DCE-MRI Improves with the Constrained Reference Region Model
Abstract
Purpose
To compare the repeatability of quantitative dynamic contrast enhanced (DCE) MRI with the reference region model (RRM) against the conventional Tofts model and to test if it can be improved using the two-step Constrained RRM (CRRM).
Methods
The models were evaluated on a publicly available DCE-MRI dataset. Ten prostate cancer patients were selected from the dataset, each having (1) two baseline exams acquired within a two-week period, (2) labeled regions-of-interest (ROIs), and (3) acceptable data quality. The datasets included ROIs for the tumor, normal peripheral zone, and whole prostate. Additional ROIs were drawn for the reference tissue (gluteus maximus) and for the arterial input function (AIF, in the femoral artery). The RRM and CRRM were fitted voxel-wise on the prostate using muscle as the reference region. For comparison purposes, the Tofts model was also fitted, once with the measured AIF in each individual patient (Tofts-IndvAIF) and another with a population-averaged AIF from literature (Tofts-PopAIF). The %Repeatability Coefficient (%RC), defined from the within-subject coefficient of variation (wCV) as 2.77*wCV*100%, was calculated for Ktrans and kep in the tumor ROI from each model.
Results
For kep, the RRM had a %RC of 979% and this improved to 45% with the CRRM, indicating higher repeatability. In comparison, the Tofts-IndvAIF led to a %RC of 84% and Tofts-PopAIF was 92%. For Ktrans, RRM had an %RC of 84%, CRRM was slightly less repeatable at 89%, Tofts-IndvAIF had 110%, and Tofts-PopAIF was 88%.
Conclusion
The CRRM substantially improved the repeatability of kep over the original RRM, while that of Ktrans remained relatively unchanged. The CRRM also showed better repeatability for kep when compared against the Tofts model, while repeatability for Ktrans was comparatively similar for all models.