Exploration of the Prognostic Value of Radiomics Analysis and Deep Learning of Coronary Plaques on Computed Tomography for Cardiovascular Events

背景和目标：冠状动脉斑块可能导致心血管事件，并且有早期证据表明，计算机断层扫描（CT）上易损斑块特征的无创检测和随后的医疗管理变化可能会改善结局。CT在疑似冠状动脉疾病（CAD）患者中的使用越来越多，并且是在一次检查中捕获所有冠状动脉节段中斑块三维信息的最佳无创成像检查。然而，现有的数据是有限的斑块形态的可变定义，在事件收集的不一致性，和混合的研究设计。此外，关于冠状动脉CT斑块特征对稳定性胸痛患者的预后价值的前瞻性数据有限。我们提出对CT上的冠状动脉斑块进行放射组学分析和深度学习（DL），以识别心血管事件高危患者。在这个项目中，我们将探讨放射组学和DL的能力，全面的评价和量化冠状动脉plaques.Methods和工作计划：我们前瞻性地收集了1700多例患者的CT图像数据在多中心的放电试验在临床相关的预后设置可疑CAD，并将进行长期的临床随访。在拟议的项目中，我们将使用卷积神经网络对冠状动脉斑块分析测试放射组学和DL的预后价值。首先，我们将测试该前瞻性患者队列中冠状动脉钙评分的预后价值。其次，我们将使用传统的斑块分割和分类（例如，非钙化、部分钙化、钙化），并将其预后价值与高风险斑块特征如低衰减、阳性重塑、餐巾环征和斑点状钙化进行比较。第三，我们将调整和实施现有的基于放射组学和DL的图像分析算法，以提取易损斑块的特征。这包括上述高风险冠状动脉斑块特征、放射组学特征和DL预测，以比较前瞻性定义的长期临床终点（如心肌梗死和冠状动脉血运重建）的预测准确性。图像分析的定量结果将作为心血管事件的数据库提供。预期的知识增益：我们预期对冠状动脉斑块特征与稳定性胸痛患者预后之间的关联有新的见解，这将加强放射组学分析和冠状动脉CT深度学习的临床意义。最终，这将允许识别易患心肌梗死的患者，并使用超过25000名无症状个体的基于人群的SCAPIS项目的数据测试高级冠状动脉CT斑块分析的有效性和普遍性，该项目将在第二个三年资助期内获得长期随访数据。

Background and Objectives: Coronary artery plaques may lead to cardiovascular events and there is early evidence that noninvasive detection of vulnerable plaque features on computed tomography (CT) and subsequent changes in medical management may lead to improved outcomes. CT is an increasingly used test in patients with suspected coronary artery disease (CAD) and is the best noninvasive imaging test to capture three-dimensional information about plaques in all coronary artery segments in one examination. However, existing data is limited by variable definitions of plaque morphology, inconsistency in event collection, and mixed study designs. Moreover, there is limited prospective data on the prognostic value of coronary CT plaque features in stable chest pain patients. We propose radiomics analysis and deep learning (DL) of coronary plaques on CT to identify patients at high-risk of cardiovascular events. In this project, we will explore the capabilities of radiomics and DL to comprehensively characterise and quantify coronary artery plaques.Methods and Work Programme: We prospectively collected CT image data from more than 1700 patients in the multicentre DISCHARGE trial in the clinically relevant prognostic setting of suspected CAD and will conduct long-term clinical follow-up. In the proposed project, we will test the prognostic value of radiomics and DL using convolutional neural networks for coronary artery plaques analysis. First, we will test the prognostic value of the coronary artery calcium score in this prospective patient cohort. Second, we will analyze all CT datasets using conventional plaque segmentation and classification (e.g., non-calcified, partially calcified, calcified) and compare the prognostic value with high-risk plaque features such as low-attenuation, positive remodelling, the napkin ring sign and spotty calcifications. Third, we will adapt and implement existing radiomics- and DL-based image-analysis algorithms to extract features of vulnerable plaques. This includes the above high-risk coronary artery plaque features, radiomic features and DL-predictions to compare the accuracy in prognostication of prospectively defined long-term clinical endpoints such as myocardial infarction and coronary revascularisation. The quantitative results of image analysis will be made available as a database with cardiovascular events.Anticipated Gain of Knowledge: We anticipate novel insights into the association of coronary plaque features with prognosis of stable chest pain patients which will strengthen the clinical implications of radiomics analysis and deep learning of coronary CT. Ultimately, this will allow identifying patients prone to suffer myocardial infarction and testing the validity and generalisability of advanced coronary CT plaque analysis using data from the population-based SCAPIS project of more than 25000 asymptomatic individuals which will have long-term follow-up data for the second three-year funding period.