Cybersecure End-to-End FPGA-Accelerated ECG Monitoring for Precision Diagnosis With Personalized CWT and Adversarial...

Area of Science:

• Biomedical Engineering
• Cybersecurity
• Edge Computing

Background:

• Wearable technology and edge computing are transforming cardiovascular monitoring.
• There is a growing need for private, secure, and real-time diagnostic solutions.
• Existing systems face challenges in balancing accuracy, privacy, and resource constraints.

Purpose of the Study:

• To present an edge-wearable ECG monitoring system with enhanced privacy and security.
• To achieve high diagnostic accuracy and real-time inference on resource-constrained hardware.
• To mitigate privacy risks and improve robustness against adversarial attacks.

Main Methods:

• Integration of personalized Continuous Wavelet Transform (CWT) for feature extraction and privacy preservation.
• Implementation of a DeepFool-FGSM adversarial defense strategy with quantization-aware training (QAT).
• Optimized parallel PoolFormer architecture deployed on Field Programmable Gate Array (FPGA).

Main Results:

• Achieved high diagnostic accuracy of 98.93% on FPGA.
• Demonstrated real-time inference with latency below 1.7 ms.
• Showcased improved robustness against adversarial perturbations with low FPGA-core power consumption (0.055 W).

Conclusions:

• The developed system offers a cybersecure, personalized ECG monitoring solution at the edge.
• It ensures confidentiality, integrity, and availability for cardiovascular diagnostics.
• The system effectively balances performance, security, and resource efficiency for wearable applications.