基于大数据的心脏性猝死预测中心电关键特征提取与分析

This project is focused on the investigation of the key issues in feature extraction and analysis of dynamic electrocardiogram (ECG) signals for forecasting the sudden cardiac death in the context of telemedicine. Considering the individual differences as well as the increased level of signal noise and interference in telemedicine environments, we will investigate a set of ECG signal enhancement and characteristic wave detection algorithms for arrhythmia classification and Q-T feature extraction based on deep learning. The main research of this project includes: (1) the de-noising method based on adaptive wavelet threshold and de-noising auto-encoder will be integrated to improve the system robustness under complex noise; (2) by analyzing the ECG signals’ characteristics in energy domain, energy window will be introduced to build R wave extraction algorithm and then Q-wave and S-wave will be calibrated based on R-wave and morphology; (3) ECG signals’ phase space will be constructed using the chaos theory, and the chaotic representation of arrhythmia signals will be obtained by calculating Lyapunov exponential, and arrhythmia classification will be realized by designing deep neural network composing of auto-encoder and softmax classifier; (4) the T wave morphology will be identified based on the deep belief networks, and then the key points will be detected guiding by the morphology character and the Q-T features will be analyzed eventually. This research has strong innovative as well as practical values. The result of this research can be directly applied to telemedicine systems.

本项目围绕心脏性猝死预测中动态心电信号特征提取和分析关键问题展开研究。充分考虑人体个体差异特征，增加此背景下信号噪声多、干扰大等因素，利用心电信号具有大数据特征的优势，引入深度学习，研究心电信号去噪和特征波检测算法，构建心律不齐分类和Q_T特征提取方法。主要研究内容包括：（1）将自适应阈值小波去噪法与降噪自动编码器方法相结合提高系统对复杂噪声的鲁棒性；（2）分析心电信号的能量域特性，利用能量窗变换构建R波提取算法，基于R波和形态特征完成Q波和S波的标定；（3）构建心电信号的相空间，通过计算Lyapunov指数得到心律不齐信号的混沌表示，采用自动编码机构造深神经网络分类器完成心律不齐分类；（4）基于深神经网络识别T波形态，然后在形态特征指导下检测关键点，分析Q-T特征。此研究成果可直接应用于远程医疗系统，具有较强的创新性和实用价值。

研究了心电信号去噪算法，将降噪自动编码深度学习网络引入到心电信号降噪中，并融合小波变换实现了多类噪声的有效去除，在保证算法精度的前提下提高了实时性。考虑导联间信号的差异性，构建了不同导联之间推导的统计模型，实现了多导联联合去噪，将信噪比提升到21.54dB，均方根误差小于0.0401。.研究了心电信号特征波自动提取算法，利用小波变换多频域分析优势，结合能量窗变换和时域修正算法，实现了R波、P波和T波提取的准确提取，三种特征波的平均检出率大于99%。提出了改进吉布斯采样与基于深度卷积神经网络的T波形态学分类算法，有效提高了T波检测精度，T波峰值的位置误差为-0.60±51ms，T波终止点的误差为-2.44±78ms。.研究了心律不齐心电信号的特征构建方法，利用混沌域表示描述心电信号，提取了心电信号Lyapunov指数曲线震荡特性作为心律不齐的混沌特征，实现了室性早搏的早期识别。构建了堆栈稀疏自动编码网络自动提取心律不齐心电信号信号深度特征，实现了6类心拍的准确识别，总体识别精度达99.5%，对噪声具有更高的鲁棒性，更符合实际系统中复杂环境下的心律不齐分类需求。.研究了心脏性猝死智能预测算法，从心电信号的时序信号出发，引入深层回声状态网络自动提取心电信号的心脏猝死特征，实现了正常和心脏性猝死的智能区分，心脏性猝死的智能预测敏感度达96.6%。扩展研究了心肌梗死自动定位与预测算法，分别利用多导联特征的串联和浅层的稀疏自动编码机实现了下壁心肌梗死和多类心肌梗死的准确定位，平均检测精度达到99.9%。

内容获取失败，请点击重试