房颤病人的心电特征研究及单导联心电信号房颤波提取方法

Automatic extraction of F-wave from ECG is the key to monitor atrial fibrillation (AF) patients around the clock and at any places. This project, centered around extracting F-wave from single-lead ECG, studies the characteristics of the atrial and ventricular action potential after the occurrence of AF and the feature extraction method based on MCA. The detailed steps include: 1) based on animal experiment results, adopting the HH model and the cellular automaton model respectively to study the characteristics of cardiac muscle cell action potential after the occurrence of AF and the transmission mechanism of ECG, and thus constructing the analytical model of atrial and ventricular activity action potential and obtaining the prior information for extracting F wave; 2) using the alternating projection algorithm and find ways to construct the equiangular tight frame dictionary with the analytical model of atrial and ventricular activity action potential and explore ways to apply the prior information in MCA through the dictionary; 3) based on the structural characteristics of the equiangular tight frame dictionary, designing a fast and stable MCA algorithm and improving the accuracy and stability of the F-wave extraction. This project aims to reveal the features of electrical signals of tissues inside the heart after the occurrence of AF and optimize the extraction of the single-lead ECG, and thus offer theoretical guidance for monitoring AF with portable medical devices.

自动化的从心电信号中提取房颤波是全天候移动式监护房颤病人的关键。本项目以单导联心电信号下的房颤波提取问题为背景，研究房颤后的心房和心室电向量特点及基于形态分量分析的特征提取方法：1）结合动物实验，分别采用Hodgkin-Huxley模型和细胞自动机模型，研究房颤后心肌细胞的动作电位特点和心电传播机制，从而建立心房和心室活动电向量的解析模型，得到用于提取房颤波的先验信息；2）采用交替投影法，研究由心房和心室活动电向量解析模型构造近似等角紧框架字典的方法，探索通过字典将先验信息用于形态分量分析法的途径；3）根据近似等紧支框架字典结构特点，设计快速、稳定的形态分量分析求解算法，优化房颤波提取的准确率和稳定性。项目旨在揭示房颤后心脏内各组织的电信号特点、优化单导联心电信号的提取方法，为移动医疗下的房颤监护提供理论指导。

(1)通过动物实验研究了房颤发生后心房内心肌细胞的动作电位特点，从而推演出了房颤心电信号的特征；通过对大量临床心电图的收集和分析，开发出了一种基于卷积神经网络算法的房颤诊断模型；将该模型部署到上海某三甲医院进行了2个多月的临床测试，在医生诊断出的55例房颤心电图中，算法诊断出54条，敏感度达到98.18%，总共3677条心电图中和医生专家一致率达到95.26%。将产品成功应用到了上海青浦的多个社区，帮助医生高效筛查出了多位隐匿性的房颤病人，使患者得到了及时治疗，生活质量得到了提高。产品在第一届世界人工智能大会展位上得到展览，参观者络绎不绝，医生们对产品也是赞不绝口。.(2)培养研究生3名，发表论文10篇，其中SCI收录3篇，EI论文2篇。获授权发明专利6项。