基于CNN结合双向LSTM网络的心电图异常分析

摘要心血管疾病是导致人类死亡或残疾的主要原因之一，具有发病率高、致残率高和并发症多的特点。人口老龄化和寿命延长使慢性疾病风险增加，尤其是心血管疾病。然而，我国基层医疗资源严重不足。为了满足人口和社会发展的需求，建立新型智慧医疗体系对慢性疾病至关重要。心电图作为心脏疾病的诊断主要手段，能反映心脏的病理特征。心电图还是监测和诊疗心血管疾病、老年痴呆和中风等慢性疾病的基础，具有非侵入性、经济性和便捷灵活性等特点。心电是一种微弱、不稳定、低频的生物电信号，且在不同个体和场景下存在很大的差异性。本文主要针对心电数据使用CNN网络，DNN网络和CNN结合双向LSTM网络来分析与识别。训练前需准备MT-BIH数据集和cpsc-2018数据集来为网络训练。在模型训练完成后，可以得到模型的混淆矩阵和1oss曲线。通过混淆矩阵可以计算出模型的准确率、灵敏度、特异度和℉1值来比较网络的性能。随着心电监护设备的轻便化和普及化，利用更多心电数据研究出一种识别种类多、识别准确率更高的心律失常辅助分类算法，具有很强的现实意义。结果表明，相对于其他网络模型，CNN结合双向LSTM网络模型的效果更好，性能更强一些。关键词：心电图：心电异常：深度学习：双向长短时记忆网络；深度残差网络：卷积神经网络1AbstractCardiovascular disease is one of the main causes of death or disability in humans,withhigh morbidity,high disability rate and many complications.Population ageing and longer lifespans increase the risk of chronic diseases,especially cardiovascular disease.However,China's primary medical resources are seriously insufficient.In order to meet the needs ofpopulation and social development,the establishment of a new smart medical system isessential for chronic diseases.As the main means of diagnosing heart diseases,electrocardiogram can reflect the pathological characteristics of the heart.ECG is also thebasis for monitoring and treating chronic diseases such as cardiovascular diseases,Alzheimer's disease and stroke,and is non-invasive,economical and flexible.ECG is a weak,unstable,low-frequency bioelectric signal,and there are greatdifferences in different individuals and scenarios.This paper mainly uses CNN network,DNNnetwork and CNN combined with bidirectional LSTM network for analysis and identificationof ECG data.Before training,you need to prepare the MIT-BIH dataset and the cpsc-2018dataset to train the network.After the model training is completed,the confusion matrix and loss curve of the modelcan be obtained.The confusion matrix can be used to calculate the accuracy,sensitivity,specificity,and FI values of the model to compare the performance of the network.With the lightness and popularization of ECG monitoring equipment,it is of strongpractical significance to use more ECG data to develop an arrhythmia-assisted classificationalgorithm with multiple identification types and higher recognition accuracy.The results showthat compared with other network models,CNN combined with bidirectional LSTM networkmodel has better effect and stronger performance.Keywords:Electrocardiogram;Abnormal ECG;Deep learning;Bidirectionallong-short-term memory network;Deep residual network;Convolutional neural network21绪论1.1研究意义及背景1.1.1研究意义心电图(Electrocardiogram,ECG),现已是现代医学的常规检查。它是一种记录心脏电活动的非侵入性检测方法，它可以记录心脏电活动，反映心脏功能和结构状态，对诊断心脏疾病有重要价值，而且它是评估心脏疾病的最安全和最有力的方法四。现代科技飞速发展，心电图的测量变得更加方便和快捷，为心电图自动分析提供了大量的信息。人口老龄化导致我国心血管病患者持续增加，心血管疾病已经成为居民的疾病死亡率的第一位，农村和城市居民分别为46.66%和43.81%四。心电图是重要的判断标准。但是心电图识别需要医师有着非常丰富的经验，而经验丰富的医师条件显然在一些乡村中是不具备的。本文实现的心电异常识别技术有一定的意义，在此处列出了以下几个方面：首先，心电异常识别技术可以提高心脏病的早期发现和及时干预的能力。心血管疾病往往在没有明显症状的情况下发生。心电异常识别技术可以通过对心电图进行实时或离线的分析，发现潜在的心脏问题，如心律失常、缺血、梗塞等，并及时提醒医生或患者采取相应措施，从而降低由心血管疾病导致的死亡比例。其次，心电异常识别技术能够提高心电图诊断的准确性和一致性。心电图诊断是一项需要高