ADVANCED CONVOLUTIONAL NEURAL NETWORKS FOR BIOLOGICAL AGE ESTIMATION FROM ECG

DOI: 10.31673/2412-4338.2026.019023

Authors

  • Володимир Георгійович Сліпченко, (Volodymyr Slipchenko) National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Kyiv, Ukraine https://orcid.org/0000-0002-3405-0781
  • Любов Григорівна Полягушко, (Liubov Poliahushko) National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Kyiv, Ukraine https://orcid.org/0000-0003-3287-8523
  • Владислав Валерійович Шатило, (Vladyslav Shatylo) National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Kyiv, Ukraine https://orcid.org/0000-0001-5395-2097
  • Дмитро Олександрович Антонкін, (Dmytro Antonkin) National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute", Kyiv, Ukraine https://orcid.org/0009-0002-9615-9077

Abstract

This study focuses on improving the efficiency and accuracy of human biological age estimation by developing an intelligent software system based on advanced deep convolutional neural network (CNN) architectures. The main emphasis is on automated electrocardiogram (ECG) analysis using data from the Jena University Hospital database (1,121 subjects). The relevance of this research stems from the fact that traditional risk markers, such as chronological age, often fail to reflect the true physiological wear and tear of the cardiovascular system, whereas heart biological age serves as a critical integrated health indicator. The study examines the limitations of classical machine learning methods based on manual feature engineering (heart rate variability, interval durations) compared to end-toend deep learning approaches that learn directly from raw signals. To address the critical issue of class imbalance in the 15-class age distribution, techniques such as SMOTE and Focal Loss were applied, along with implementing a patientlevel data splitting strategy to ensure test set independence.

The study results significantly improved biological age classification accuracy. While classical models (Logistic

Regression, Random Forest) achieved a maximum accuracy of 49%, and Recurrent Neural Networks (Bi-LSTM) reached

64.4%, the proposed Advanced CNN architecture, enhanced with an Attention mechanism and residual connections (ResNet), achieved a classification accuracy of over 87% with a mean absolute error of 2.6 years. The model demonstrated high sensitivity even for underrepresented geriatric groups (91+ years), achieving a recall above 70%. The results provided important insights into deep learning's capability to detect hidden non-linear aging patterns in raw ECG signals, such as QRS complex fragmentation and conduction velocity changes, which are inaccessible to visual interpretation or linear analysis. The developed "BioAge-ECG AI" software system can be used for mass screening and early detection of cardiovascular risks.

Keywords: Biological Age, ECG, Deep Learning, CNN, ResNet, Attention Mechanism, Class Imbalance.

References

  1. Belsky, D. W., et al. (2015). Quantification of biological aging in young adults. Proceedings of the National Academy of Sciences, 112(30), E4104-E4110. https://doi.org/10.1073/pnas.1506264112
  2. Lima, E. M., et al. (2021). Deep neural network-estimated electrocardiographic age as a mortality predictor. Nature Communications, 12(1), 5117. https://doi.org/10.1038/s41467-021-25351-7
  3. Attia, Z. I., et al. (2019). Age and sex estimation using artificial intelligence from standard 12-lead electrocardiograms. Circulation: Arrhythmia and Electrophysiology, 12(9), e007284. https://doi.org/10.1161/CIRCEP.119.007284
  4. Kiranyaz, S., et al. (2021). 1D Convolutional Neural Networks and Applications: A Survey. Mechanical Systems and Signal Processing, 151, 107398. https://doi.org/10.1016/j.ymssp.2020.107398
  5. Schumann, A., & Bär, K. J. (2022). Autonomic function regulating blood pressure and cardiac rhythm in aging and disease. Scientific Data, 9(1), 1-10. https://doi.org/10.1038/s41597-022-01121-5
  6. Goldberger, A. L., et al. (2000). PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation, 101(23), e215-e220. https://doi.org/10.1161/01.CIR.101.23.e215
  7. Welch, P. (1967). The use of Fast Fourier Transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms. IEEE Transactions on Audio and Electroacoustics, 15(2), 70-73. https://doi.org/10.1109/TAU.1967.1161901
  8. Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: synthetic minority oversampling technique. Journal of artificial intelligence research, 16, 321-357. https://doi.org/10.1613/jair.953
  9. Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11), 2673-2681. https://doi.org/10.1109/78.650093
  10. He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition, 770-778. https://doi.org/10.1109/CVPR.2016.90
  11. Vaswani, A., et al. (2017). Attention is all you need. Advances in neural information processing systems, 30. https://proceedings.neurips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
  12. Kingma, D. P., & Ba, J. (2014). Adam: A Method for Stochastic Optimization. arXiv preprint arXiv:1412.6980. https://doi.org/10.48550/arXiv.1412.6980
  13. Lin, T. Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017). Focal loss for dense object detection. Proceedings      of         the        IEEE    international     conference        on computer          vision, 2980-2988. https://doi.org/10.1109/ICCV.2017.324

Downloads

Published

2026-04-03

Issue

No. 1 (2026)

Section

Articles