基于动态心电信号深度神经网络分析的阻塞性睡眠呼吸暂停早期预警研究

阻塞性睡眠呼吸暂停(Obstructive sleep apnea，OSA）的持续存在和发展可引起多器官多系统损害。但OSA早期症状隐匿易被忽视，因此早期预警至关重要。尽管动态心电信号被广泛用于OSA的筛查预测，但现有技术不能全面准确辨识OSA相关心电特征，因而无法达到精准预警的目的。根据我们前期研究成果提示：新兴的深度卷积神经网络（CNN）技术在克服上述瓶颈问题，实现OSA早期精准预测方面具有潜在价值。本研究在课题组前期构建OSA患者睡眠监测和动态心电图数据库的基础上，采用GAN-LSTM对心电噪声进行建模和降噪，基于GAN-CNN检测心电波形，利用CNN技术对正常人及OSA患者睡眠心电信号进行全面系统分析，并进一步验证，从而为OSA早期精准预警和病情预测提供简便可靠的新方法，同时为利用先进人工智能成果突破疾病医学诊断的瓶颈问题拓展思路。

The continued existence and development of obstructive sleep apnea (OSA) can cause multiple damages of organs and multiple systems. However，the early symptoms of OSA are concealment and easy to be overlooked, so the early warning of OSA is essential. Although dynamic ECG signals are widely used in the screening and prediction of OSA, the technology used cannot fully and accurately identify OSA-related ECG characteristics, and thus cannot achieve the purpose of accurate warning. According to our previous research results, the emerging deep convolutional neural network (CNN) technology has potential value in overcoming the above bottlenecks and achieving early accurate prediction of OSA. In this study, based on the early construction of the sleep monitoring and dynamic ECG database of OSA patients，GAN-LSTM is used to model and reduce the ECG noise, GAN-CNN is used to detect ECG waveform, CNN is used to conduct a comprehensive system analysis of sleep ECG signals in normal people and OSA patients, after that we will further verify the reults in clinical. This study will provide a simple and reliable new method for early accurate warning and prediction of OSA. At the same time, which will expand the ideas for using advanced artificial intelligence to break through the bottleneck of diagnosis of diseases.