Quantification Of Carotid Atherosclerotic Plaques In Contrast Enhanced Ultrasound Images

Atherosclerosis, a type of cardiovascular and cerebrovascular disease, is a major cause of mortality in modern societies. If the vulnerable plaques rupture, they will cause local thrombosis and embolism, leading to stroke. Recent studies have shown that the vulnerability of plaques is related with the intraplaque neovascularization. The neovascularization encourages the ongoing process of atherosclerosis, resulting in plaque rupture. Contrast enhanced ultrasound (CEUS) is a noninvasive imaging technique that is able to detect the intraplaque neovascularization, and it provides a new way to assess the atherosclerotic carotid lesions.The technology of contrast enhanced ultrasound develops fast. However, the medical image processing and analysis methods, such as image filtering, segmentation and feature extraction, can not meet users’demands, and they still stay in a stage of qualitative grading. In order to overcome the disadvantages of qualitative analysis, we mainly focus on algorithms of image filtering, segmentation and feature extraction for contrast enhanced ultrasound, to provide computer-aided quantification of plaques in CEUS images.First, speckle noise is an inherent property of ultrasound images, so we perform image filtering before subsequent processing. Anisotropic diffusion (AD) is a classic noise reduction method, but the gradient edge detector in AD can not distinguish edges and noise, limiting its robustness. In this paper, we introduce McIlhagga edge detector, a more robust edge detector, into AD model to improve its performance. Experimental results show that when the variance of noise is low (<0.1),the performance of the proposed method is comparable to that of the traditional methods; while the variance increases to0.2-0.5, the peak signal to noise ratio, Pratt’s figure of merit, structural similarity index values of our methods increase by8.38%,302.45%and9.80%respectively, compared with the traditional methods.Second, image segmentation of CEUS images is a crucial technique for quantification of plaques. There are two steps, including the segmentation of plaques and the segmentation of intraplaque neovascularization. For the segmentation of plaques, we propose a new method based on spatial-temporal analysis and snakes. For the segmentation of intraplaque neovascularization, three different thresholding methods are performed on a single frame and an accumulated image respectively. Experimental results of plaque segmentation show that the proposed method outperforms the traditional method, by0.06mm,2.0%,0.04mm, and5.3%, in terms of the mean distance error, relative mean distance error, mean signed distance error, and relative difference degree, respectively. Experimental results of intraplaque neovascularization segmentation demonstrate that the multilevel thresholding on the accumulated image is the best option for segmentation among six different combinations of methods.Third, feature extraction is adopted for quantification of plaques in CEUS images. In this paper, two categories of features are extracted, i.e.,morphological features and texture features. Morphological features include area ratio, center deviation degree and sparsity. Texture features consist of first-order statistics and features from the gray level co-occurrence matrix. In total, four morphological features and32texture features are extracted from33cases of carotid plaques in24patients. Then the statistical analysis is carried out to find the relationship between the quantitative features and qualitative grading. The results show that in a system of two grades, i.e.,the high grade and the low grade, there are10parameters presenting significant difference between the two grades (t-test, P<0.05); in a system of three grades, there are five parameters exhibiting significant difference between the three grades (analysis of variance, P<0.05).