Liquid State Machine-Based Electrocardiogram Classification System: Challenges, Design Trade-Offs, and Opportunities
IEEE Access · 2026 · 14 · 115747-115770
DOI: 10.1109/access.2026.3717654 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Ricardo Perez-Lopez und 5 weiteren Autor:innen. Die Arbeit nennt 91 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Secretariat of Science, Humanities, Technology, and Innovation (SECIHTI) of Mexico through the Scholarship.
Abstract (Originalsprache)
In recent years, artificial intelligence systems for arrhythmia detection in electrocardiogram (ECG) signals have achieved remarkable performance, supporting the development of automated tools for clinical time-series analysis. However, traditional machine learning approaches usually rely on handcrafted features, reducing inherently spatio-temporal signals to static representations. Although deep learning architectures can overcome this limitation by learning representations directly from data, they often entail high computational costs, large training requirements, and complex optimization procedures. These limitations motivate the exploration of alternative computational paradigms capable of preserving temporal information while reducing training complexity. Liquid State Machines (LSMs), a paradigm within Reservoir Computing (RC), offer such an alternative by projecting temporal inputs into dynamic, high-dimensional representations through a recurrent, task-independent liquid or reservoir, while learning is confined to a simple task-dependent readout. This study investigates the design choices in an LSM-based ECG heartbeat classification system, emphasizing its pipeline components and the practical challenges of applying RC-based models to biomedical time-series data. A comprehensive experimental analysis of the LSM-based pipeline is conducted, including signal preprocessing, encoding strategies, and readout configurations. Beyond performance evaluation, this work examines how these pipeline components influence the behavior, consistency, and practical trade-offs of LSM-based ECG classification. The results provide insights into the opportunities and limitations of LSM architectures for clinical time-series processing, contributing to the understanding of RC methods for physiological signal analysis.
Förderung
- Secretariat of Science, Humanities, Technology, and Innovation (SECIHTI) of Mexico through the Scholarship
Zitieren
APA
Perez-Lopez, R., Ibarra-Manzano, O. G., Rostro-Gonzalez, H., Lara-Perez, J. I., Sotelo-Figueroa, M. A., & Espinal, A. (2026). Liquid State Machine-Based Electrocardiogram Classification System: Challenges, Design Trade-Offs, and Opportunities. IEEE Access, 14, 115747-115770. https://doi.org/10.1109/access.2026.3717654
Vancouver
Perez-Lopez R, Ibarra-Manzano OG, Rostro-Gonzalez H, Lara-Perez JI, Sotelo-Figueroa MA, Espinal A. Liquid State Machine-Based Electrocardiogram Classification System: Challenges, Design Trade-Offs, and Opportunities. IEEE Access. 2026;14:115747-115770. doi:10.1109/access.2026.3717654
BibTeX
@article{PerezLopez2026,
author = {Perez-Lopez, Ricardo and Ibarra-Manzano, Oscar G. and Rostro-Gonzalez, Horacio and Lara-Perez, Jesus Ignacio and Sotelo-Figueroa, Marco A. and Espinal, Andres},
title = {Liquid State Machine-Based Electrocardiogram Classification System: Challenges, Design Trade-Offs, and Opportunities},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {115747--115770},
doi = {10.1109/access.2026.3717654}
}
EUVIMED