一种面向CTA图像的冠状动脉自动分割算法

An Automatic Segmentation of Coronary Artery with CTA

为了实现冠状动脉自动分割并提高分割精度,本文结合卷积神经网络和改进的区域生长算法,提出了一种自动分割算法.该算法主要包括三个部分:种子点的定位,血管相似性特征的提取和基于相似性特征的区域生长.首先利用卷积神经网络分割升主动脉,为后续的生长提供种子点;然后计算血管相似性特征,为后续的区域生长提供约束特征;最后,结合血管相似性特征和数据的密度信息进行区域生长.本文的分割结果比传统的区域生长方法的分割效果更好,与手动标注的平均体素误差为1.983 2 HU/体素.

In this paper,a method of automatic segmentation of coronary artery is proposed based on convolutional neural network and improved region growth algorithm. This method has three important parts:seeds detection,vascular similarity feature extraction and regional growing. First,the ascending aorta is segmented with a convolutional neural network to provide seed points for subsequent growing. Then,vascular similarity feature is extracted to provide features for subsequent region growing. Finally,region growing is performed by fusing the similarity characteristic of the vessels and the density information of the data. The segmentation result is better than that of the traditional regional growth method,and the mean voxel error with manual annotation is 1. 983 2 HU/voxel.