Arterial Wall Motion Estimation in Carotid Artery Using Deep Learning with Extended Kalman Filter

Arterial wall motion estimation of the carotid artery is known to be extremely useful in the diagnosis of atherosclerotic diseases. While several approaches have been developed for robust tracking of the arterial wall, real-Time tracking for hand-held devices remains a challenging task. Towards this end, we propose a deep learning-based tracker with a non-linear motion model describing the dynamics of the arterial wall. The proposed model uses a Siamese architecture-based convolutional neural network for detection and a non-linear motion model that incorporates an Extended Kalman Filter. Initial experiments proved that the proposed model works in near real-Time with promising tracking accuracy when compared with conventional techniques such as exhaustive search-based similarity matching.