Cardiac MR-Based Risk Stratification for Heart Failure and Atrial Fibrillation in HCM

肥厚型心肌病(HCM)是最常见的遗传性心脏病，影响多达 总人口中有1：200人。HCM，最初在心脏性猝死(SCD)的背景下描述， 通常与心力衰竭(HF)和房颤(AF)有关。过去两年的严谨研究 几十年来，我们能够识别出患有SCD风险最高的HCM患者，他们可以受益于 预防性植入式心律转复除颤器(ICD)。随着SCD预防、HCM管理的进步 现已将重点转向高频和自动对焦。近50%的肥厚性心肌炎患者有轻度到重度的心力衰竭症状。高频是 现在被认为是与HCM相关的死亡的最常见原因。房颤是最常见的持续性房颤 心律失常，发生在近25%的肥厚性心肌病患者中，并导致生活质量下降和 中风风险增加。目前，我们无法预测哪些肥厚型心肌炎患者更有可能进展为 晚期心力衰竭或发展为房颤。使用超声心动图和心脏磁共振的心血管成像已经发挥了 在我们对HCM不断发展的理解中发挥核心作用。超声心动图对左心功能作出可靠的评估 心脏(LV)流出道梗阻和舒张期功能障碍。它的高空间分辨率和显著的 组织定征能力，心脏磁共振已经成为一种非常适合定征的成像方式 HCM表型。这项建议的目标是通过杠杆作用开发新的风险分层范例 人工智能(AI)在改善HCM患者管理方面的最新进展。我们将调查一个深层次的问题 用于预测心血管不良结局的学习(DL)风险模型，该模型包含(A)标准临床 和成像参数以及(B)使用(I)放射组学分析(即 自动提取和选择临床有意义的成像标记的计算方法)或(Ii)深度 使用深度卷积神经网络(CNN)提取的图像特征。这些产品的性能 模型将使用在塔夫茨医学中心、BIDMC和 多伦多大学。

Hypertrophic cardiomyopathy (HCM) is the most common genetic heart disease, affecting as many as 1:200 individuals in the general population. HCM, initially described in the context of sudden cardiac death (SCD), is commonly associated with heart failure (HF) and atrial fibrillation (AF). Rigorous research over the past two decades has enabled us to identify HCM patients at the greatest risk of SCD who could benefit from a prophylactic implantable cardioverter defibrillator (ICD). With advances in SCD prevention, HCM management has now shifted its focus to HF and AF. Nearly 50% of HCM patients have mild to severe HF symptoms. HF is now considered the most common cause of HCM-related mortality. AF is the most common sustained arrhythmia, occurring in nearly 25% of HCM patients, and responsible for a decreased quality of life and increased stroke risk. Currently, we are not able to predict which HCM patients are more likely to progress toward end-stage HF or develop AF. Cardiovascular imaging using echocardiography and cardiac MR has played a central role in our evolving understanding of HCM. Echocardiography provides a robust assessment of left ventricular (LV) outflow obstruction and diastolic dysfunction. With its high spatial resolution and remarkable tissue characterization capabilities, cardiac MR has emerged as an imaging modality well suited to characterize the HCM phenotype. The goal of this proposal is to develop novel risk stratification paradigms by leveraging recent advances in artificial intelligence (AI) to improve HCM patient management. We will investigate a deep learning (DL) risk model for prediction of adverse cardiovascular outcomes that incorporates (a) standard clinical and imaging parameters and (b) novel cardiac MR signatures extracted using (i) radiomic analysis (i.e. a computational method to automatically extract and select clinically significant imaging markers) or (ii) deep imaging signatures, extracted using deep convolutional neural networks (CNN). The performance of these models will be rigorously evaluated using 3 HCM cohorts collected at Tufts Medical Center, BIDMC, and University of Toronto.