Detection of Dangerous Arrhythmias Using a 1D Convolutional Neural Network Trained on the Full Spectrum of a Short ECG Signal

An essential practical task is the timely recognition of dangerous arrhythmias since the early detection of this pathology can prevent mortality and improve the quality of life. The article considers a solution to the problem of heart rhythm disturbances binary classification using the 1-Dimensional Convolutional Neural Network. The work focused on the detection of dangerous ventricular arrhythmias because of their severe consequences. We used the full spectrum of two-second fragments of the electrocardiogram for model training. 1D CNN architecture for sequence classification allows learning directly on raw data while showing satisfactory results. The model achieved an overall test accuracy of 95.7 percent with sensitivity and specificity of 97.7% and 90.2%, respectively.