Cine phase-contrast MR flow measurements: improved precision using an automated method of vessel detection.

Abstract

The purpose of this study was to construct a method of vessel edge detection that correctly identifies vessel pixels and to compare the interuser variability of cine phase contrast MR volumetric flow rates obtained with the conventional manual method and an automated method. The automated method was developed based on a magnitude image threshold value and compared with the manual method in a flow phantom (three users) and in velocity images of the portal vein (five users). The threshold value determined from the magnitude image was applied to a region of interest surrounding the vessel of interest on the magnitude image to construct a vessel edge detection mask m(x,y). The velocity images were then multiplied by the mask m(x,y) and volumetric flow rates determined using the identified vessel pixels. In the flow phantom, flow measurements with the magnitude threshold method had significantly less interuser variability compared with the manual method (p < 0.01) and were within 10% (mean 6.0%) of the actual flow versus 35% (mean 18.6%) with the manual method. Regarding flow measurements in the portal vein of six volunteers, the magnitude threshold method was significantly more precise (p < 0.01) than the manual method with a mean standard deviation between the five users of 40.4 +/- 12.9 ml/min (range 22-60 ml/min) and 110.4 +/- 32.7 ml/min (range 70-155 ml/min), respectively. The magnitude threshold method of vessel edge detection developed in this study yields flow measurements that are accurate in the model system and have significantly less interuser variability than the manual method. This method shows promise for improving the precision of cine phase contrast flow measurements.