面向肠系膜上动脉栓塞的自动识别算法

An Automated Recognition Algorithm for Mesenteric Arterial Embolism Detection

急性肠系膜上动脉栓塞是一种死亡率高的疾病,通过计算机断层扫描血管造影可以提供明确的诊断。目前主要靠临床医生的观察诊断,没有计算机辅助诊断肠系膜上动脉栓塞自动识别。因此本文提出“腹部动脉分割—分支标注—栓塞识别”方法,首先采用改进的区域生长法对腹部动脉自动分割,实现腹部血管的精确分割。然后采用图匹配算法,结合图像位置信息和几何特征实现分支标注,实现从腹部动脉说中准确提取肠系膜上动脉分支。最后建立灰度共生矩阵等提取图像的一阶,二阶纹理特征,通过极致梯度提升树(eXtreme Gradient Boosting,XGBoost)算法分类器实现栓塞识别。实验结果表明在血管分割阶段本实验血管脉络和细节特征更加清晰,栓塞识别平均精确度可达0.939,在不同类型的肠系膜上动脉栓塞样本中均有较好的识别效果。

Acute superior mesenteric artery thrombosis is a high-mortality disease computed tomography angiography provides a definitive diagnosis.Currently,clinical observation is the primary diagnostic method,as there is no computer-aided diagnosis system for automatically identifying superior mesenteric artery embolism.Therefore,this paper proposes the“Abdominal Artery Segmentation—Branch Annotation—Embolism Recognition”approach.First,an improved region growing method is used for automatic abdominal artery segmentation,achieving precise abdominal vascular segmentation.Then a graph matching algorithm is used to realize branch annotation by combining image position information and geometric features to accurately extract the branches of the superior mesenteric artery from the abdominal artery theory.Finally,a grayscale co-occurrence matrix is established to extract the first-order and second-order texture features of the image,and embolism identification is realized through the eXtreme Gradient Boosting(XGBoost)algorithm classifier.The experimental results show that in the vascular segmentation stage,the vascular networks and detailed features of this experiment are clearer,the average accuracy of embolism recognition can reach 0.939,and the recognition effect is good in different types of superior mesenteric artery embolism samples.