A study of coronary artery rotational motion with dense scale-space optical flow in intravascular ultrasound

This paper describes a novel method for estimating tissue motion in two-dimensional intravascular ultrasound (IVUS) images of a coronary artery. It is based on the classical Lukas-Kanade (LK) algorithm for optical flow (OF). The OF vector field quantifies the amount of misalignment between two consecutive frames in a sequence of images. From the theoretical standpoint, two fundamental improvements are proposed in this paper. First, using a simplified representation of the vessel wall as a medium with randomly distributed scatterers, it was shown that the OF equation satisfies the integral brightness conservation law. Second, a scale-space embedding for the OF equation was derived under the assumption of spatial consistency in IVUS acquisitions. The spatial coherence is equivalent to a locally affine motion model. The latter effectively captures and appropriately describes a complex deformation pattern of the coronary vessel wall under the varying physiological conditions (i.e. pulsatile blood pressure). The accuracy of OF tracking was estimated on the tissue-mimicking phantoms subjected to the controlled amount of angular deviation. Moreover, the performance of the classical LK and proposed approach was compared using the simulated IVUS images with an atherosclerotic lesion. The experimental results showed robust and reliable performance of up to 5 degrees of rotation, which is within the plausible range of circumferential displacement of the coronary arteries. Subsequently, the algorithm was used to analyze vessel wall motion in 18 IVUS pullbacks from 16 patients. The in vivo experiments revealed that the motion of coronary arteries is primarily determined by the cardiac contraction.