S. Sali1, Z.M. Yuldashev2

1,2 St. Petersburg Electrotechnical University "LETI" (St. Petersburg, Russia)

1 salar.sali97@gmail.com, 2 zmyuldashev@etu.ru

The development of automated systems for analyzing electrocardiograms (ECG) requires reliable methods for quickly identifying clinically significant arrhythmias from short signal fragments. Simultaneous classification of a large number of arrhythmia classes is a complex task, which justifies the use of a hierarchical approach with sequential identification of the main rhythm groups and subsequent specialized classification of ventricular arrhythmias, including ventricular fibrillation (VF).

Objective of the work – development and evaluation of a two-stage method for classifying short ECG fragments to automatically detect ventricular arrhythmias and ventricular fibrillation.

A scientific justification has been provided for the use of a hierarchical approach to the classification of short ECG fragments. A software pipeline has been developed and studied, combining four PhysioNet MIT-BIH databases, unified signal preprocessing, ECG segmentation into overlapping five-second fragments, hierarchical automatic labeling, and a hybrid one-dimensional residual convolutional neural network. In the first stage, the fragments are classified into normal rhythm (Normal), atrial arrhythmia (Atrial Arrhythmia), and ventricular arrhythmia (Ventricular Arrhythmia); in the second stage, fragments without VF (Non-VF) and ventricular fibrillation (Ventricular Fibrillation) are distinguished. According to the results of the final assessment, the accuracy was 0.875 for the first stage, 0.870 for the second stage, and 0.914 for the entire hierarchical system. The ROC AUC values were 0.940, 0.952, and 0.964, respectively.

The two-stage system proposed and described in the paper makes it possible to reduce the complexity of multi-class ECG classification by sequentially determining the rhythm group and specifically detecting ventricular fibrillation using short signal fragments. The obtained results can be used in the development of automated ECG analysis tools and decision support systems for identifying dangerous ventricular arrhythmias.

The obtained technical characteristics and formulated recommendations for the use of the system can be applied in the preparation of an experimental layout for subsequent clinical testing with the participation of blind and visually impaired users.

Sali S., Yuldashev Z.M. Two-stage classification of dangerous arrhythmias over short time intervals // Biomedical Radioelectronics. 2026. V. 29. № 6. P. 85−91. DOI: https://doi.org/10.18127/j15604136-202606-15

1. Zeppenfeld K., Tfelt-Hansen J., de Riva M. et al. 2022 ESC Guidelines for the management of patients with ventricular arrhythmias and the prevention of sudden cardiac death. European Heart Journal. 2022. V. 43(40). P. 3997–4126. DOI:10.1093/eurheartj/ehac262.
2. Xiao Q., Lee K., Mokhtar S.A., Ismail I., Pauzi A.L.B.M., Zhang Q., Lim P.Y. Deep Learning-Based ECG Arrhythmia Classification: A Systematic Review. Applied Sciences. 2023. V. 13(8). P. 4964. DOI:10.3390/app13084964.
3. Ansari Y.A., Mourad O., Qaraqe K.Q., Serpedin E. Deep learning for ECG Arrhythmia detection and classification: an overview of progress for period 2017–2023. Frontiers in Physiology. 2023. V. 14. P. 1246746. DOI:10.3389/fphys.2023.1246746.
4. Ribeiro A.H., Ribeiro M.H., Paixão G.M.M. et al. Automatic diagnosis of the 12-lead ECG using a deep neural network. Nature Communications. 2020. V. 11. P. 1760. DOI:10.1038/s41467-020-15432-4.
5. Hannun A.Y., Rajpurkar P., Haghpanahi M. et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nature Medicine. 2019. V. 25. P. 65–69. DOI:10.1038/s41591-018-0268-3.
6. Moody G.B., Mark R.G. MIT-BIH Arrhythmia Database. PhysioNet. Version 1.0.0.
7. Moody G.B., Mark R.G. MIT-BIH Atrial Fibrillation Database. PhysioNet. Version 1.0.0.
8. Greenwald S.D. MIT-BIH Supraventricular Arrhythmia Database. PhysioNet. Version 1.0.0.
9. Albrecht P., Moody G.B., Mark R.G. MIT-BIH Malignant Ventricular Ectopy Database. PhysioNet. Version 1.0.0.
10. Niu J., Tang Y., Sun Z., Zhang W. Inter-Patient ECG Classification With Symbolic Representations and Multi-Perspective Convolutional Neural Networks. IEEE Journal of Biomedical and Health Informatics. 2020. V. 24(5). P. 1321–1332. DOI:10.1109/JBHI.2019.2942938.