A Hybrid Protection Scheme for the Gait Analysis in Early Dementia Recognition

Area of Science:

• Biometrics and Human Activity Recognition
• Neurodegenerative Disease Diagnostics
• Data Security and Privacy

Background:

• Gait abnormalities are key indicators for early detection of neurodegenerative diseases like dementia.
• Biometric data, including gait patterns, require robust privacy and security measures during processing.
• Existing methods for gait analysis may not adequately address privacy concerns.

Purpose of the Study:

• To propose and evaluate an innovative hybrid protection scheme for gait features.
• To ensure patient privacy in human activity recognition (HAR) systems for early dementia diagnosis.
• To integrate the protection scheme with a long short-term memory (LSTM) neural network for a secure diagnostic system.

Main Methods:

• Developed a hybrid protection scheme combining partially homomorphic encryption and cancelable biometrics via random projection.
• Implemented the scheme with a long short-term memory (LSTM) neural network for secure early dementia diagnosis.
• Ensured compliance with privacy standards like ISO/IEC 24745.

Main Results:

• The proposed scheme achieved a high trade-off between data security and system performance.
• Accuracy degradation was minimal, at most 1.20%, compared to unprotected systems.
• The protection scheme demonstrated scalability and independence from specific neural network architectures.

Conclusions:

• The hybrid protection scheme effectively secures gait features while enabling accurate early dementia detection.
• The system offers a scalable and privacy-preserving solution for HAR in healthcare.
• The approach provides a strong foundation for secure biometric data processing in sensitive applications.