EUVIMEDEUVIMED entdecken
Crossref

FedSmartCare: Design and Implementation of a Federated Learning Enabled Vital-Sign Monitoring System

Bithi Banik, Yuan Lin, Andrii Shalaginov, Debasish Ghose

IEEE Sensors Journal · 01.05.2026 · 26 (9) · 13948-13967

DOI: 10.1109/jsen.2026.3676888 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.05.2026. Verfasst von Bithi Banik und 3 weiteren Autor:innen. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 60 Literaturangaben. Gefördert von: UTFORSK Programme, Norwegian Directorate for Higher Education and Skills, Sustainable AI Literacy in Higher Education through Multilateral Collaborations.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The Medical Internet of Things (MIoT) is revolutionizing healthcare through continuous and real-time monitoring of vital-signs, significantly benefiting applications ranging from general wellness and chronic disease management to elderly care. Integrating artificial intelligence (AI) into vital-sign analysis has further advanced disease detection and predictive healthcare capabilities. However, conventional AI approaches often require transmitting sensitive health data collected by MIoT devices to centralized servers for model training and analysis, which raises substantial privacy and security concerns. To address these challenges, this study introduces FedSmartCare, a modular MIoT platform integrated with federated learning (FL), designed explicitly for real-time, privacy-preserving monitoring and management of vital-signs. By employing FL paradigms, FedSmartCare facilitates collaborative, decentralized AI model training at edge devices, significantly reducing privacy risks associated with centralized data handling. The proposed system’s performance is comprehensively validated through rigorous experiments on real-world data collected via FedSmartCare, along with three publicly available datasets. Multiple advanced AI models are systematically evaluated to assess the platform’s accuracy, efficiency, and robustness for classification and forecasting tasks involving both homogeneous and heterogeneous FL settings. In forecasting tasks, long short-term memory (LSTM) obtained mean absolute error (MAE) values of approximately 8.2 and 3.5 BPM under homogeneous clients across two distinct datasets, while gated recurrent unit (GRU) achieved MAE values of 9.5 BPM under heterogeneous clients after 20 training rounds. For stress classification, a residual network (ResNet)-based model reached 99% accuracy in the homogeneous setting over the same training horizon. These results demonstrate that FedSmartCare offers strong analytical performance while preserving data privacy, highlighting its potential for secure and intelligent MIoT-based healthcare solutions.

Förderung

  • UTFORSK Programme
  • Norwegian Directorate for Higher Education and Skills
  • Sustainable AI Literacy in Higher Education through Multilateral Collaborations

Zitieren

APA

Banik, B., Lin, Y., Shalaginov, A., & Ghose, D. (2026). FedSmartCare: Design and Implementation of a Federated Learning Enabled Vital-Sign Monitoring System. IEEE Sensors Journal, 26(9), 13948-13967. https://doi.org/10.1109/jsen.2026.3676888

Vancouver

Banik B, Lin Y, Shalaginov A, Ghose D. FedSmartCare: Design and Implementation of a Federated Learning Enabled Vital-Sign Monitoring System. IEEE Sensors Journal. 2026;26(9):13948-13967. doi:10.1109/jsen.2026.3676888

BibTeX
@article{Banik2026,
  author = {Banik, Bithi and Lin, Yuan and Shalaginov, Andrii and Ghose, Debasish},
  title = {FedSmartCare: Design and Implementation of a Federated Learning Enabled Vital-Sign Monitoring System},
  journal = {IEEE Sensors Journal},
  year = {2026},
  volume = {26},
  number = {9},
  pages = {13948--13967},
  doi = {10.1109/jsen.2026.3676888}
}
Kostenlos starten →