DRFAS-enhanced personalized federated learning for secure and robust cardiovascular disease detection
- 1School of Computer Science and Engineering, Vellore Institute of Technology, Chennai Campus
- 2School of Electronics and Engineering, Vellore Institute of Technology, Chennai Campus
Abstract
The integration of privacy protection in smart healthcare systems has revolutionized disease detection and patient monitoring, but privacy concerns arise from the collection and analysis of sensitive health data. Federated learning (FL) offers data privacy to diverse clients, but a single global model fails. Personalized FL addresses privacy concerns, focusing on individual needs. The proposed differential privacy personalized federated learning with a deep learning model (Per-FLDL), along with Distributed Robust Federated Aggregation Service (DRFAS) techniques, accurately detects, predicts, and improves correlation understanding of cardiovascular disease in smart healthcare systems, thereby allowing the clients to design their own privacy-preserving personalized FL models. The proposed system minimizes data exposure, enhances local model training, and improves personalization and illness detection accuracy by utilizing the FL approach. By utilizing federated learning, decentralized model training is enabled while guaranteeing that personal health data remains on-device, reducing dangers associated with data breaches and illegal access. The proposed Per-FLDL framework achieved high accuracy rates of 98.79%, 98.05%, and 99.32% in healthcare, heart disease, and cardiovascular disease datasets, respectively.
Related
- Causality-driven hybrid intrusion detection with entropy-aware feature learning and federated intelligence
- Personalized Prevention of Cardiovascular Diseases Through Measurement of Plaque Burden in the Carotid Artery
- Evolving practices in pretransplant cardiovascular screening: moving toward personalized assessments
- SGLT2 Inhibitors and Cardiovascular Outcomes in Patients With Fabry Disease: Observations From a Federated Research Network
- Pharmacogenomics-guided cardiovascular therapy in low- and middle-income countries: A systematic review of barriers, equity gaps, and pathways to personalized medicine
BibTeX
@article{karthik2026drfasenhanced,
title = {{DRFAS-enhanced personalized federated learning for secure and robust cardiovascular disease detection}},
author = {V. Karthik and OmKumar ChandraUmakantham and Sudhakaran Gajendran and Suguna Marappan},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-55738-9},
url = {https://doi.org/10.1038/s41598-026-55738-9},
note = {CC BY},
}
RIS
TY - JOUR
TI - DRFAS-enhanced personalized federated learning for secure and robust cardiovascular disease detection
AU - V. Karthik
AU - OmKumar ChandraUmakantham
AU - Sudhakaran Gajendran
AU - Suguna Marappan
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-55738-9
UR - https://doi.org/10.1038/s41598-026-55738-9
AB - The integration of privacy protection in smart healthcare systems has revolutionized disease detection and patient monitoring, but privacy concerns arise from the collection and analysis of sensitive health data. Federated learning (FL) offers data privacy to diverse clients, but a single global model fails. Personalized FL addresses privacy concerns, focusing on individual needs. The proposed differential privacy personalized federated learning with a deep learning model (Per-FLDL), along with Distributed Robust Federated Aggregation Service (DRFAS) techniques, accurately detects, predicts, and improves correlation understanding of cardiovascular disease in smart healthcare systems, thereby allowing the clients to design their own privacy-preserving personalized FL models. The proposed system minimizes data exposure, enhances local model training, and improves personalization and illness detection accuracy by utilizing the FL approach. By utilizing federated learning, decentralized model training is enabled while guaranteeing that personal health data remains on-device, reducing dangers associated with data breaches and illegal access. The proposed Per-FLDL framework achieved high accuracy rates of 98.79%, 98.05%, and 99.32% in healthcare, heart disease, and cardiovascular disease datasets, respectively.
SN - 2045-2322
ER -