Motion-compensated frame rate up-conversion in carotid ultrasound images using optical flow and manifold learning.

Abstract

Carotid ultrasonography is a reliable and non-invasive method to evaluate atherosclerosis disease and its complications. B-mode cineloops are widely used to assess the severity of atherosclerosis and its progression; ho- wever, tracking rapid wall motions of the carotid artery is still a challenging issue due the low frame rate. The aim of this paper was to present a new hybrid frame rate up-conversion (FRUC) method that accounts for motion based on manifold learning and optical flow. In the last decade, manifold learning technique has been used to pseudo-increase the frame rate of carotid ultrasound images, but due to the dependence of this method to the number of recorded cardiac cycles and frames, a new hybrid method based on manifold learning and optical flow was proposed in this paper. Locally linear embedding (LLE) algorithm was first applied to find the relation between the frames of consecutive cardiac cycles in a low dimensional manifold. Then by applying the optical flow motion estimation algorithm, a motion compensated frame was reconstructed. Consequently, a cycle with more frames was created to provide a more accurate consideration of carotid wall motion compared to the typical B-mode ultrasound ima-ges. The results revealed that our new hybrid method outperforms the pseudo-increasing frame rate scheme based on manifold learning.