Machine learning-based biomechanical analysis for thoracic aortic aneurysm rupture risk assessment

项目摘要 在美国，主动脉瘤疾病一直排在前20位死亡原因之列。 人口胸主动脉瘤（TAA）是成人死亡的主要原因。进展 胸主动脉瘤是一个无声的过程，但破裂/剥离可能突然发生，这往往导致死亡。 这些致命的事件是可以通过选择性手术修复来预防的， 干预措施指出，当胸主动脉瘤最大直径达到5至10 mm时，应进行手术。 5.5厘米但是，该标准无法评估较小TAA（直径≤ 5 cm）的风险。是 据估计，美国有数百万TAA患者的TAA较小，这些患者 不幸的是，目前的标准忽略了这一点。因此，在这个项目中，我们提出了一个创新的 整合机器学习（ML）和计算生物力学的风险方法 评估较小的TAA。为了实现这一目标，我们将开发（1）ML模型， 从3D临床CT图像重建胸主动脉几何形状，这将使快速和 TAA风险的简化分析，（2）实时TAA压力分析的ML模型，以及（3） 融合测量和计算的患者特征的概率风险指数（例如， 几何形状、应力、材料强度等），并考虑到来自不同 源将在现有的1000个数据集上开发和验证所提出的方法 患者和从600例患者的纵向随访研究中收集的新数据集， 这将是第一个大规模的基于机器学习的TAA生物力学分析研究 风险评估这项研究将导致心血管领域的突破 计算建模和应用机器学习，为如何更好地 评估TAA风险，并减少TAA疾病的沉默和突然杀手造成的死亡。

PROJECT SUMMARY Aortic aneurysm disease ranks consistently in the top 20 causes of death in the U.S. population. Thoracic aortic aneurysm (TAA) is a leading cause of death in adults. The progression of TAA is a silent process, yet rupture/dissection can occur suddenly, which often causes death. The deadly events are preventable by elective surgical repair, and the current criterion for surgical intervention states that surgery should be performed when TAA maximum diameter reaches 5 to 5.5 cm. However, this criterion cannot assess the risk of smaller TAAs (diameter≤5cm). It is estimated that there are millions of TAA patients in the U.S. with smaller TAAs, and these patients are unfortunately ignored by the current criterion. Thus, in this project, we propose an innovative approach of integrating machine learning (ML) and computational biomechanics for risk assessment of smaller TAAs. To achieve this goal, we will develop (1) ML models for automated thoracic aorta geometry reconstruction from 3D clinical CT images, which will enable a fast and streamlined analysis of TAA risk, (2) ML models for realtime TAA stress analysis, and (3) a probabilistic risk index that fuses the measured and computed patient characteristics (e.g. geometry, stress, material strength, etc) and takes into account uncertainties from different sources. The proposed approach will be developed and validated on an existing dataset of 1000 patients and a new dataset to be assembled from a longitudinal follow-up study of 600 patients, which will be the first large-scale study of machine learning-based biomechanical analysis for TAA risk assessment. This study will lead to a breakthrough in the fields of cardiovascular computational modeling and applied machine learning, provide new insights on how to better assess TAA risk, and reduce death by the silent and sudden killer of TAA disease.