The Research For Heart Block Location With Convolutional Neural Network

Objective: Heart block is a common cardiovascular disease.Severe heart block can threaten human life.It is convenient and efficient to diagnose heart block diseases by identifying ECG,which is a common method for clinical diagnosis of heart block.Convolutional neural networks have been applied to image and speech recognition.The study on ECG identification of heart block has great clinical significance.In this study,an algorithm was designed to accurately identify the precise location of heart block.Methods: The core algorithm based on convolution neural network is proposed in this study: Multi-resolution Connected Res Net,with the hospital ECG data and MIT-BIH arrhythmia database for experimental data source,extracted from training set and test set used to train and test network respectively,and compared the Multi-resolution Connected Res Net with classical convolution neural networks.The visualization and statistical analysis of the training result were to evaluate the performance of the Multi-resolution Connected Res Net.Results: 1.Compared with the classical convolutional neural network,the Res Net with resolution connection converges faster and the convergence curve is smoother.2.The Res Net training clinical database with resolution connection has an accuracy rate of92.4%,a specificity of 98%,a sensitivity of 92.4%,an F1 score of 0.93,and an AUC of 0.98.The accuracy,specificity,sensitivity,F1 score and AUC of the training mit-bih database were 97.7%,99.2%,97.4%,0.97 and 0.98,these were higher than the classical convolutional neural network.3.The Res Net with resolution band connection can clearly identify heart block with a high degree of differentiation.Conclusion: Multi-resolution Connected Res Net performs well in the recognition task of accurate positioning for cardiac block,with an accuracy rate of over 90%.Multi-resolution band connected Res Net can be used in the auxiliary diagnosis of clinical cardiac block.