Method for calculating confidence intervals for phase contrast flow measurements

BACKGROUND Phase contrast (PC) measurements play an important role in several cardiovascular magnetic resonance (CMR) protocols but considerable variation is observed in such measurements. Part of this variation stems from the propagation of thermal noise from the measurement data through the image reconstruction to the region of interest analysis used in flow measurement, which limits the precision. The purpose of this study was to develop a method for direct estimation of the variation caused by thermal noise and to validate this method in phantom and in vivo data. METHODS The estimation of confidence intervals in flow measurements is complicated by noise correlation among the image pixels and cardiac phases. This correlation is caused by sequence and reconstruction parameters. A method for the calculation of the standard deviation of region of interest measurements was adapted and expanded to accommodate typical clinical PC measurements and the region-of-interest analysis used for such measurements. This included the dependency between cardiac phases that arises due to retrospective cardiac gating used in such studies. The proposed method enables calculation of standard deviations of flow measurements without the need for repeated experiments or repeated reconstructions. The method was compared to repeated trials in phantom measurements and pseudo replica reconstructions of in vivo data. Three different flow protocols (free breathing and breath hold with various accelerations) were compared in terms of the confidence interval ranges caused by thermal noise in the measurement data. RESULTS Using the proposed method it was possible to accurately predict confidence intervals for flow measurements. The method was in good agreement with repeated measurements in phantom experiments and there was also good agreement with confidence intervals predicted by pseudo replica reconstructions in both phantom and in vivo data. The proposed method was used to demonstrate that the variation in cardiac output caused by thermal noise is on the order of 1% in clinically used free breathing protocols, and on the order of 3-5% in breath-hold protocols with higher parallel imaging factors. CONCLUSIONS It is possible to calculate confidence intervals for Cartesian PC contrast flow measurements directly without the need for time-consuming pseudo replica reconstructions.