Measurement of vascular diameter in vitro by automated software for CT angiography: effects of inner diameter, density of contrast medium, and convolution kernel.

Abstract

This investigation was performed to evaluate the accuracy of diameter measurement of vessels in vitro by automated software for CT angiography. Vascular models with three inner diameters ( approximately 3, 4, and 6 mm) filled with contrast medium of three different densities ( approximately 460, 350, and 210 H) were scanned with helical CT. Five convolution kernels (soft, standard, detail, bone, and lung) were used. We evaluated the measurement error, defined as the difference between the diameter measured by the automated software and the true inner diameter of the vascular model. Statistical analysis involved three-way analysis of variance with repeated measures. Significant differences occurred in measurement error among the three vascular model inner diameters, among the three densities of intravascular contrast medium, and among the five convolution kernels (p < 0.01). In all the convolution kernels except lung, measurement errors progressively decreased with higher densities of intravascular contrast medium (p < 0.01). In vascular models filled with contrast medium of 350 H, measurement errors were significantly smaller in soft (mean +/- standard deviation [SD], 0.29 +/- 0.16 mm) and bone (0.23 +/- 0.05 mm) than in other convolution kernels (p < 0.01). The accuracy of diameter measurement was affected by the vascular model inner diameter, the density of contrast medium, and the convolution kernel. A higher density of intravascular contrast medium and selection of the proper convolution kernel will improve accuracy.