基于神经网络深度学习的全心冠状动脉斑块定量化分析及破裂风险预警模型研究

冠状动脉粥样硬化易损斑块具有症状隐匿性、破裂突发性的特点，一旦破裂将导致急性冠脉事件（ACS）。目前对易损斑块的识别主要是依据有创性冠脉腔内影像，尚缺乏精准的早期无创性影像评价手段。申请人及团队前期纵向研究提出基于冠状动脉CTA（CCTA）的易损斑块特征可独立预测ACS的发生，研究结果被纳入美国SCCT/ACR/ACC冠状动脉CAD-RADS报告专家共识。然而，这些易损斑块特征判读受到肉眼识别局限性和医生经验差异性的限制。本项目拟通过人工智能卷积神经网络技术对CCTA数据进行深度挖掘，自动提取易损斑块潜在的定量化特征信息，建立易损斑块高精度识别系统；在此基础上，将获得的CCTA易损斑块自动识别方法与基于深度学习的CT实时血流动力学分析技术相整合，创建更加高效、灵敏、精准的无创性斑块破裂风险预警模型，为未来冠状动脉斑块破裂的智能化无创精准检测和早期预警提供全面可靠的依据和关键性的技术手段。

The vulnerable coronary plaque is characterized by latent symptoms and potential risk of sudden rupture. The current major approach for vulnerable plaque assessment is invasive intra-coronary imaging and the effective non-invasive imaging modality for vulnerable plaque early detection is absent. Our previous longitudinal study has demonstrated that high risk plaque features detected by coronary CTA are independent predictors for acute cardiovascular events, which has been cited by Coronary Artery Disease - Reporting and Data System (CAD-RADS): An Expert Consensus Document of SCCT, ACR and ACC. However, the identification of those features are limited by visual analysis and individual experiences. The current proposal aims to comprehensively extract potential quantitative high risk plaque characteristics from coronary CTA by the artificial intelligence technology of the convolution neural network and therefore to establish accurate automatic vulnerable plaque identification system. Moreover, with the integration of accurate automatic vulnerable plaque identification methods and deep learning based CT hemodynamic analysis approach, a comprehensive and effective non-invasive risk stratification model for plaque rupture will be established. With the above functional and anatomical evaluation of vulnerable plaque by CTA based artificial intelligence technology, a key technique for non-invasive accurate detection and early prediction of coronary plaque rupture will be provided.