Arrhythmia classification based on convolution neural network feature extraction and fusion

This study proposes a new automatic classification method of arrhythmias to assist doctors in diagnosing and treating arrhythmias. The convolution neural network is constructed to extract the features of ECG signals and wavelet components of QRS complex. The ECG signal features and wavelet features extracted by the network and the artificially extracted RR interval features are input to the full connection layer for fusion, and the softmax function is used to classify the beats in the output layer. The network is trained and tested using the mil lead data in MIT BIH arrhythmia database. The overall classification accuracy of this method is 98.12%, the average sensitivity is 87.32%, and the average positive predictive value is 90.37%. This method can quickly identify different types of arrhythmias, and has certain reference value for the application of computer-aided diagnosis of arrhythmias