超声图像中基于经验模态分解自动提取内–中膜厚度

Objective  This study introduces an improved empirical mode decomposition (EMD) algorithm for measuring intima medial thickness (IMT) based on image segmentation techniques, in order to improve the accuracy and stability of traditional manual measurement of IMT. Methods  An improved EMD algorithm is proposed for noise reduction, in which the control lattice densities of the bi-dimensional B-splines adopted in the interpolation step are different in horizontal and vertical orientations. The modification is based on the fact that, in the original ultrasound images, the requirements of precision and smoothness in two directions are different. K-means algorithm is then used to separate the lumen, vessel wall and other structures, followed by mathematical morphological methods used for obtaining the final segmentation of intima media. Results  The improved EMD algorithm performed well in denoising and reconstructing the image, which effectively overcomed the challenges in carotid ultrasound image segmentation with due to heavy noise and low-resolution. The results showed that the proposed segmentation method had achieved a good accuracy witch much lower complexity comparing to the methods based on deformable models. Conclusion  The improved EMD algorithm can effectively remove the ultrasonic noise while preserving details and edges of the structure, and therefore has great potential for automatic extraction of IMT in carotid ultrasound images. It should also be applicable to extract stripe structures under heavy noise environment according to the principles of the method.