冠状动脉疾病是目前世界范围内导致死亡的首位疾病，该疾病通常由血管狭窄引起，而各类血管斑块是血管狭窄的主要成因，因此对冠状动脉血管斑块的识别与分类以及对血管狭窄的临床功能评估具有重要意义。本项目围绕冠状动脉疾病诊断在血管分割、斑块识别以及狭窄功能评估等方面存在的挑战性问题，开展基于CT血管造影成像的冠脉斑块识别及狭窄功能评估方法研究。研究融合先验知识的Graph Cut 分割方法，实现对冠状动脉血管网络的三维快速准确分割；在高性能GPU 计算平台上采用深度卷积网络对冠状动脉血管斑块的形态学特征进行学习，实现对复杂血管斑块的识别与分类；以斑块引起的血管狭窄的多种特征为输入，以实际测得的血流储备分数为输出，建立随机森林回归分析模型，利用该模型对血管斑块引起的冠脉狭窄进行临床功能评估，判断其引起心肌缺血的情况。本项目为冠状动脉疾病的精准诊断和手术规划提供新的方法和手段。

Coronary artery disease is currently the leading cause of death worldwide, which is usually caused by vascular stenosis. On the other hand, a variety of vascular plaques take responsibility for the coronary artery stenosis. Therefore, the recognition and classification of coronary artery plaque and the clinical functional assessment of coronary artery stenosis are of great importance. Focusing on the diagnostic challenges of coronary artery disease on vascular segmentation, plaque recognition and stenosis function assessment, the project conducts a study on coronary plaque recognition and stenosis function assessment based on computed tomography angiography. Graph Cuts method incorporated with prior knowledge is studied to implement the three-dimensional fast and precise segmentation of coronary artery networks. The deep convolutional neural network is utilized to learn the morphological features of the vascular plaques on high-performance GPU computing platforms, which realizes the recognition and classification of complex vascular plaques. The Random Forest regression model is built by taking various features of vascular stenosis as the input, and taking real fractional flow reserve values as the output. Utilizing the Random Forest regression model, the project conducts the clinical functional assessment of coronary artery senosis caused by vascular plaques, so as to judge the severity of myocardial ischemia. The project provides new method and technique for the precise diagnosis and surgery planning of coronary artery disease.