Study On Intelligent Classification Of Arrhythmia Detection In 12-lead Electrocardiogram

Cardiovascular diseases are a serious threat to human life and health,ranking as the first cause of death from non-communicable diseases worldwide.ECG is a non-invasive and eco-nomical diagnostic tool for arrhythmias widely used in clinical practice.ECG monitoring system plays an important role in the prevention,detection and treatment of cardiovascular diseases with convenient and efficient diagnosis and treatment service mode,and how to achieve high effi-ciency and high precision intelligent analysis of ECG signal by ECG signal is the main challenge now.The traditional ECG analysis methods have low diagnostic efficiency,poor adaptability and low level of intelligence,which cannot meet the analysis needs of ECG signal diversity and variability characteristics.The main work and innovations made in this dissertation on the analysis process of 12-lead ECG in ECG monitoring system are as follows:1.In this dissertation,a temporal convolutional network（E-TCN）classification model is proposed for the problems of low accuracy and complex model of intelligent classification model for arrhythmia detection in 12-lead ECG.Firstly,the perceptual field and depth are designed,and a temporal convolutional network model structure is proposed.Secondly,the temporal convolutional network can extract the spatial features within each ECG lead,and the causality between the convolutional layers can deal with the temporal features between different leads,and the residual neural network solves the degradation problem of the deep network.The tem-poral convolutional network performs well in a twelve-lead ECG arrhythmia classification task and is experimentally validated on the China Physiological Signaling Challenge 2018（CPSC-2018）dataset,where an F₁score of 83.6%is obtained with only 400,000 counts,and the model is validated by plotting its ROC curve on each classification and calculating the AUC area.performance.2.In this dissertation,we designed a multi-feature fusion classification model（ME-TCN）for 12-lead ECG based on a temporal convolutional network model for classification of arrhyth-mia data in 12-lead ECG,incorporating multiple dimensional features such as personal features and heart beat features,for the practical application scenarios of ECG intelligent classification systems in clinical settings.From clinical studies to data analysis,correlations between per-sonal features such as age and gender and arrhythmia-like diseases were found,and then QRS wave clusters were localized,and then the intrinsic connections of these features were uncov-ered by embedding layers.In addition,the internal features of the ECG were extracted using an autoencoder,and finally these features were fused with the features extracted by a temporal convolutional network for classification,combining individual patient differences with the in-ternal features of the ECG and improving the model’s performance on the classification task of the twelve-lead ECG dataset.Experiments were also conducted on the CPSC-2018 dataset to validate the gain of individual features and heartbeat features on the model,increasing the F₁score of the model to 85.4%,and comparing the model’s performance on each classification by the ROC curves on each classification as well as the AUC values,allowing the model to im-prove its generalization ability while making the model more compatible with the actual clinical diagnostic needs.The 12-lead ECG temporal convolutional network classification model and the multi-feature fusion classification model proposed in this dissertation improve the classification accuracy with a lower number of parameters and fuse multiple features with the model to effectively improve the efficiency of medical workers in a more practical way,enhance the intelligence of ECG monitoring system,and make cardiovascular diseases more timely and effective to be detected and treated in the field of intelligent medicine.It is of great significance in the field of intelligent medical.