Arrhythmia Classification Using 2D-CNN Models

Electrocardiogram (ECG) is one of the main tools to diagnose arrhythmia. The accurate identification of ECG signal can not only help doctors make better diagnosis, but also prevent the occurrence of cardiovascular disease. However, the current arrhythmia classification algorithms often need to be based on a large number of data sets, which reduces the scalability and practical significance of the classification algorithm. Our research proposes to use Siamese neural network based on 2D-CNN to extract the features of two-dimensional ECG signals. By calculating the Contrastive Loss and training the feature extraction model, we can judge the category of arrhythmia. The experimental results show that compared with other methods, this method not only has simple network structure, but also can be trained with fewer samples. For the five types of arrhythmias with fewer samples, the average accuracy is 97.13%.