Combination of the Level-Set Methods with the Contourlet Transform for the Segmentation of the IVUS Images.
Combination of the Level-Set Methods with the Contourlet Transform for the Segmentation of the IVUS Images.
复制标题
DOI：
10.1155/2012/439597
复制
发表时间：
2012
影响因子：
7.6
通讯作者：
Naceur MS
中科院分区：
文献类型：
--
作者：
Lazrag H;Naceur MS
血管内超声 (IVUS) 成像是一种基于导管的医学方法，已成为研究动脉粥样硬化的有用模式。 IVUS 图像中管腔和中膜外膜边界的检测是实现动脉粥样硬化可靠定量诊断的重要一步。在本文中，提出了一种自动检测管腔和中膜外膜边界的新方案。该分割方法基于水平集模型和轮廓波多分辨率分析。轮廓波变换将原始图像分解为低通分量和带通方向带。在低通带中采用圆形霍夫变换（CHT）来产生初始管腔和中膜外膜轮廓。然后，在带通子带中使用各向异性扩散滤波来抑制噪声并保留动脉边缘。最后，使用水平集函数中的曲线演化来获得最终轮廓。该方法通过 20 个模拟图像和 30 个来自人体冠状动脉的真实图像进行了实验评估。结果表明，与最近的传统水平集模型相比，平均距离误差和相对平均距离误差分别增加了5.30个像素和7.45%。这些结果表明，该方法可以自动、准确地提取两个血管边界。
Intravascular ultrasound (IVUS) imaging is a catheter-based medical methodology establishing itself as a useful modality for studying atherosclerosis. The detection of lumen and media-adventitia boundaries in IVUS images constitutes an essential step towards the reliable quantitative diagnosis of atherosclerosis. In this paper, a novel scheme is proposed to automatically detect lumen and media-adventitia borders. This segmentation method is based on the level-set model and the contourlet multiresolution analysis. The contourlet transform decomposes the original image into low-pass components and band-pass directional bands. The circular hough transform (CHT) is adopted in low-pass bands to yield the initial lumen and media-adventitia contours. The anisotropic diffusion filtering is then used in band-pass subbands to suppress noise and preserve arterial edges. Finally, the curve evolution in the level-set functions is used to obtain final contours. The proposed method is experimentally evaluated via 20 simulated images and 30 real images from human coronary arteries. It is demonstrated that the mean distance error and the relative mean distance error have increased by 5.30 pixels and 7.45%, respectively, as compared with those of a recently traditional level-set model. These results reveal that the proposed method can automatically and accurately extract two vascular boundaries.