Secure Tracking of Patient's Vital Signs Using CSI-Based Homomorphic Encryption-Enabled Deep Learning Framework.

Area of Science:

• Digital Health
• Biomedical Engineering
• Cybersecurity

Background:

• Patient privacy is a critical challenge in digital healthcare.
• Non-intrusive monitoring systems require robust data security.
• Existing methods often struggle to balance privacy with real-time processing.

Purpose of the Study:

• To introduce VitalCrypt, a novel framework for secure, real-time vital sign monitoring.
• To demonstrate the efficacy of combining Channel State Information (CSI) with homomorphic encryption and deep learning.
• To address the need for privacy-preserving solutions in healthcare monitoring.

Main Methods:

• Utilized Channel State Information (CSI) for non-intrusive monitoring.
• Implemented homomorphic encryption for direct computation on encrypted data.
• Employed signal preprocessing (Hampel, Savitzky-Golay, elliptic filters, PCA) and Power Spectral Density (PSD) feature extraction.
• Developed a lightweight neural network with encryption-compatible activation functions for classification.

Main Results:

• Achieved 99.46% accuracy for breathing rate classification on plain data and 99.44% on encrypted data.
• Demonstrated negligible performance degradation despite increased runtime due to encryption.
• Showcased effective classification of breathing and heart rates while maintaining homomorphic encryption compatibility.

Conclusions:

• VitalCrypt offers a scalable, privacy-preserving solution for real-time healthcare monitoring.
• The framework balances high accuracy with substantial data confidentiality.
• It presents a promising approach for secure deep learning applications in remote monitoring and clinical diagnostics.