Comparison of bolus arrival time determination methods for the analysis of cerebral hemodynamics by estimating the impact of temporal resolution of 4D MRA imaging

Abstract

In case of an early diagnosed cerebral vascular disease exact knowledge of the individual anatomy and hemodynamic situation is needed for improved rating of the disease and therapy planning. Concentration time curves derived from 4D magnetic resonance angiography imaging are widely used to calculate hemodynamic parameters such as the bolus arrival time (BAT). The focus of this study is to evaluate the impact of the temporal resolution to three different BAT estimation methods: model independent Time-to-Peak determination, model dependent gamma variate fit and reference based curve fitting. The three BAT estimation methods were applied to the raw, binomial smoothed and B-Spline approximated concentration time curves to evaluate a possible benefit of previous noise reduction. Additionally it is investigated whether arterial and venous structures have to be distinguished for temporal resolution requirements. For this reason 60 concentration time curves from arterial and venous structures of medical interest were extracted from six 4D TREAT MRA datasets. In a first step bolus arrival times were estimated based on the raw, binomial smoothed and B-Spline approximated curves using the three methods and the original temporal resolution. In the following these bolus arrival times served as the ground truth. Then the temporal resolution was iteratively reduced, the bolus arrival times were estimated and the relative error compared to the initial values was calculated. The results show that a minimum temporal resolution of 2.6 seconds is necessary to yield adequate results for clinical tasks. Furthermore we found that no distinction between arteries and veins has to be made. Also previous smoothing of the concentration time curves leads to higher temporal errors. The results suggest that model dependent approaches perform better compared to the model independent method whereas the reference based curve fitting yields the most robust and best results.