Page 76 - AIH-1-4

Artificial Intelligence in Health ML models for heartbeat classification

noise injection, alongside Pearson correlation and p-values, Methodology: Marc Nshimiyimana, Jovial Niyogisubizo
to enhance the classification performance to meet the level Writing – original draft: All authors
of 12-lead ECG analysis, which is considered the most Writing – review & editing: All authors
accurate analysis in clinical settings. Notably, the XGBoost
method demonstrated exceptional performance, achieving Ethics approval and consent to participate
high accuracy in heartbeat classification. Conversely, NB Not applicable.
displayed suboptimal classification capabilities among
the investigated models. The conclusions of this study are Consent for publication
summarized as follows: Not applicable.
1. Introducing FT-based feature extraction and
Gaussian noise regularization substantially improved Availability of data
the performance and robustness of ECG heartbeat
classification models. The code and data related to this study, along with more
2. The findings provided valuable insights into the technical details, can be found here: https://github.com/
comparative performance of various ML algorithms, jovialniyo93/heartbeat-classification-with-machine-models.
assisting researchers and clinicians in selecting Further disclosure
the most appropriate model for specific healthcare
applications. We confirm that this work is the result of a collaborative
3. The FT technique enabled effective capture of effort among three researchers, offering diverse perspectives,
frequency-domain information that is critical for including AI in health and the application of ML to real-
accurate heartbeat classification, thereby enhancing world problems. The contributions have been significant,
the models’ diagnostic capabilities. enhancing expertise in the field and building on our previous
4. Controlled Gaussian noise injection during training publication on predicting red wine quality using novel
proved beneficial for model generalization in real- ML methods. The first author (M.N.) is a Master’s student
world scenarios. at the Southeast University with expertise in applying
5. Our findings demonstrated the potential of ML AI, particularly ML algorithms to civil and geotechnical
in advancing cardiac healthcare monitoring and engineering fields. The third author, who is a Ph.D. student
classification, offering practical tools for more accurate at the Southeast University specializing in infrastructure
and reliable ECG-based diagnoses. resilience, project management, and risk management
(J.D.N.), contributed his perspectives to the present work.
The study’s primary limitations are the requirements
of advanced deep learning models and larger and diverse References
datasets to improve ECG heartbeat classification accuracy
and robustness. To address these limitations, future 1. Periyaswamy T, Balasubramanian M. Ambulatory cardiac
research should prioritize the exploration of deep learning bio-signals: From mirage to clinical reality through a decade
of progress. Int J Med Inform. 2019;130:103928.
architectures and the acquisition of more comprehensive
and varied datasets, which will ensure the reliability and doi: 10.1016/j.ijmedinf.2019.07.007
real-world applicability of the model in clinical settings. 2. Siontis KC, Noseworthy PA, Attia ZI, Friedman PA. Artificial
intelligence-enhanced electrocardiography in cardiovascular
Acknowledgments disease management. Nat Rev Cardiol. 2021;18(7):465-478.
None. doi: 10.1038/s41569-020-00503-2

Funding 3. Goudis CA, Konstantinidis AK, Ntalas IV, Korantzopoulos P.
Electrocardiographic abnormalities and cardiac arrhythmias
None. in chronic obstructive pulmonary disease. Int J Cardiol.
2015;199:264-273.
Conflict of interest doi: 10.1016/j.ijcard.2015.06.096
The authors declare no conflicts of interest. 4. Teymouri N, Mesbah S, Navabian SMH, et al. ECG frequency
changes in potassium disorders: A narrative review. Am J
Author contributions Cardiovasc Dis. 2022;12(3):112.
Conceptualization: Jovial Niyogisubizo 5. Faruk N, Abdulkarim A, Emmanuel I, et al. A comprehensive
Formal analysis: Jovial Niyogisubizo, Marc Nshimiyimana survey on low-cost ECG acquisition systems: Advances
Investigation: Marc Nshimiyimana, Jovial Niyogisubizo on design specifications, challenges and future direction.

Volume 1 Issue 4 (2024) 70 doi: 10.36922/aih.3543