Deep Neural Networks To Treat Atrial Fibrillation

项目总结 房颤是影响美国500多万人的一个主要健康问题，导致严重的 发病率甚至死亡率。对这种流行病的治疗是次优的，1年的成功率为30%-70% 大多数治疗方法。尽管在理解潜在的房颤机制方面取得了很大进展，但这些见解并没有 但转化为更好的房颤疗法。 该项目的科学重点集中在为异种人识别新表型的问题上 目前属于房颤范畴的情况。机器学习是一种非常适合于识别 从传统上难以分离的大型不同数据集中进行新颖的分类。我要用取款机 学习和计算方法，以分析详细的临床，结构，心脏电生理和 房颤患者的生化特征，以更好地预测各种治疗的有效和无效。 这可能使前瞻性指导能够量身定做个性化治疗。在执行这个项目时，我将成长为一名 内科医生兼科学家专注于以患者为中心的房颤研究。 该科学项目的具体目标如下：首先，我将为房颤创建一个新的疾病分类法 这对成功接受危险因素修改、抗心律失常药物治疗的患者进行了分类，或不同 消融的方法，使用计算方法和对来自我的大量训练数据的监督学习 合作者。我将在一个测试队列中评估这些疾病分区的预测效果 转诊治疗房颤。其次，我将在机器学习和患者层面使用先进的技术 分析解释为什么某种策略在个人身上可能失败或成功，为临床应用铺平道路。 第三，在一个试验性的前瞻性临床研究中，我将评估这些机器学习的可行性和准确性。 模特们。 这些实验的结果可能会通过提供房颤治疗而立即产生临床影响 以患者特定的方式进行选择，在降低风险的同时优化收益。此外，在平衡的情况下 和专家导师提供的这一奖项，我将获得必要的计算模型，临床 研究设计和生物统计学方法经验，以设计综合研究和BE 对独立资金的竞争。

PROJECT SUMMARY Atrial fibrillation (AF) is a major health problem affecting over 5 million people in the US leading to significant morbidity and even mortality. Therapy for this epidemic is suboptimal, with success of 30-70% at 1 year for most therapies. Despite great advances in understanding potential AF mechanisms, these insights have not yet translated into better AF therapy. The scientific focus of the project centers on the issue of identifying novel phenotypes for the heterogeneous conditions that currently fall under the rubric of AF. Machine learning is an approach well-suited to identify novel classifications from large diverse data sets that are traditionally difficult to separate. I will use machine learning and computational methods to analyze detailed clinical, structural, cardiac electrophysiological and biochemical features in patients with AF, to better predict responders and non-responders to various therapies. This may enable prospective guidance to tailor personalized therapy. In performing this project, I will grow as a physician-scientist focused on patient-oriented research in atrial fibrillation. The specific aims of the scientific project are as follows: First, I will create a novel disease taxonomy for AF that classifies patients successfully treated by risk factor modification, antiarrhythmic drug therapy, or diverse approaches to ablation, using computational methods and supervised learning on large training data from my collaborators. I will assess the predictive efficacy of these disease partitions in a testing cohort of patients referred for treatment of AF. Second, I will use advanced techniques in machine learning and patient-level analyses to explain why a certain strategy may fail or succeed in an individual, paving the way for clinical use. Third, in a pilot prospective clinical study, I will assess the feasibility and accuracy of these machine learning models. The findings from these experiments may provide an immediate clinical impact by delivering AF therapy options in a patient-specific manner that optimizes benefit while reducing risk. In addition, under the balanced and expert mentorship provided by this award, I will gain the necessary computational modelling, clinical research design and biostatistical methodology experience to design comprehensive studies and be competitive for independent funding.