Automatical Intima-Media Border Segmentation on Ultrasound Image Sequences Using a Kalman Filter Snake

Abstract

Segmentation of carotid intima-media (IM) borders from ultrasound images is of great importance for predicting cardiovascular risks. In this paper, we have developed a fully automatic approach to sequentially segment the carotid IM borders in each image throughout ultrasound sequences. First, the first frame of an ultrasound sequence is automatically segmented using edge detectors and dynamic programming, and then the rest frames are segmented successively under the state-space framework. Under this framework, we developed a variant of the snake method for a precise measurement. The evaluation of our segmentation result is done by comparison with average manual delineations of three physicians on a total of 65 sequences. The accuracy of our method is high. (Segmentation error is $32.1~\pm ~37.5~\mu \text{m}$ for LI and $35.0~\pm ~41.5~{\mu }\text{m}$ for MA.) The BA plot and the linear regression also demonstrate that our method is in agreement with the ground truth. This paper strengthens the potential of the state-space and snake-based approach in segmenting IM borders for clinical diagnosis by demonstrating a fully automatic scheme.