Deep learning for single-lead ECG beat arrhythmia-type detection using novel iris spectrogram representation

This paper presents a new deep learning methodology to detect among up to 17 classes of cardiac arrhythmia based on beat-wise electrocardiography (ECG) signal analysis using iris spectrogram. Automatic analysis of each ECG heartbeat makes it possible to detect abnormalities. The main aim of this paper is to develop a fast deep learning and yet an efficient approach to classify cardiac arrhythmias. The approach is implemented using 744 ECG signals for 45 persons. The approach is based on analyzing a single ECG beat and calculating the iris spectrogram. Then the iris spectrogram is fed to a convolutional neural network. The proposed method is efficient, simple and fast, which makes it feasible for real-time classification. The results show that the proposed methodology has an overall recognition accuracy of 99.13% ± 0.25, 98.223% ± 0.85, and 97.494% ± 1.26 for 13, 15, and 17 arrhythmia classes, respectively. The training/testing is performed using tenfold cross-validation. When compared to existing studies, our method is promising, outperforms many others, and can be deployed on mobile devices.