Real-time arrhythmia heart disease detection system using CNN architecture based various optimizers-networks

The main objective of this paper is to develop an interactive classifier aided deep learning system to assist cardiologists for heart arrhythmia disease classification as it shows a health-threatening condition that can lead to heart-related complications. Therefore, automatic arrhythmia heart disease detection in an early stage is of high interest as it helps to reduce the mortality rate of cardiac disease patients. In this context, a deep learning architecture is propounded for automatic classification of the patient`s electrocardiogram (ECG) signal into a specific class according to the ANSI–AAMI standards. Our proposed methodology is a multistage technique. The first stage combines an R–R peak extraction with a low pass filter applied on the ECG raw data for noise removal. The proposed second stage is a convolutional neural network (CNN) based Fully Connected layers architecture, using different networks optimizer. Different ECG databases have been used for validation purposes. The whole system is implemented on CPU and GPU for complexity analysis. For the predicted improved PTB dataset, the classification accuracy results achieve 99.37%, 99.15%, and 99.31% for training, validation, and testing, respectively. Besides, for the MIT-BIH database, the training, validation, and testing accuracies are 99.5%, 99.06%, and 99.34%, respectively. A top F1-score of 0.99 is obtained. Experimental results show a high achievement compared to the state-of-the-art models. The implementation on GPU confirms the low computational complexity of the system and the possible use in detecting disease events in real-time, which makes it a good candidate for portable healthcare devices.