Investigation of Inter-Patient, Intra-Patient, and Patient-Specific Based Training in Deep Learning for Classification of Heartbeat Arrhythmia

Effective diagnosis of electrocardiogram (ECG) is one of the simplest and fastest ways to assess the heart's function. In the recent decade, various attempts have been made to automate the classification of electrocardiogram signals to detect heartbeat arrhythmias based on deep learning. However, due to the lack of a comprehensive standard for how to divide the database into the train and test datasets and the variety of methods used for this purpose, it is not possible to make a fair comparison between many of these studies. One of the main criteria for creating train and test datasets that have a great impact on the final results is their distribution paradigm. There are three paradigms for this purpose, including Inter-Patient, Intra-Patient, and Patient-Specific. In this research, we have conducted a detailed study of the impact of these three paradigms on the final results obtained from a CNN-based deep learning model for the classification of heartbeat arrhythmia into five classes. The experimental results on the standard arrhythmia dataset show that the Patient-Specific reached the best average performance in all of the metrics. Also, this training pattern is more practical and can be employed to create patient customized devices for the classification of ECG arrhythmia.