A novel method to generate synthetic ultrasound data of the carotid artery based on in vivo observation as a tool to validate algorithm accuracy

Ultrasound imaging represents a well designed modality to estimate the motion of biological tissues in vivo, from which relevant clinical information can be assessed. However, the lack of ground truth constitutes a challenging issue when it comes to evaluate the accuracy of computerized methods. Indeed, quantification of the reliability of experimental results often involves manual or visual human operations, which may introduce subjectivity and variability. Nonetheless, numerical simulation of the imaged tissues allow a comparison with a known reference. For this purpose, we propose in this work a realistic kinematic multi-layer model of the common carotid artery. A set of 10 models was generated by randomly positioning scatterers, on which intensity, specular reflection, and bidimensional motion over the duration of one cardiac cycle were applied. Two computerized methods, namely a block-matching method and a segmentation method, were also applied on our model using identical parameter settings as those used for in vivo clinical data, in the objective to assess their accuracy. The tracking errors were 42 ± 40 μm and 12 ± 10 μm in the longitudinal and radial directions, respectively. The segmentation errors were 28±18 μm for the lumen diameter, and 15±10 μm for the intima-media thickness. We conclude from these results that our model can constitute a reliable method to quantify the accuracy of computerized algorithms.