Trip Detection Algorithms for Healthy and Amputee Individuals.

Area of Science:

• Biomechanics
• Prosthetics
• Robotics

Background:

• Amputation significantly impairs gait biomechanics, increasing fall risk due to reduced strength, balance, and toe clearance.
• Falls in amputees lead to physical injuries and psychological distress.
• Powered and microprocessor prosthetic legs require advanced algorithms for detecting external perturbations to enhance safety.

Purpose of the Study:

• To develop and validate novel algorithms for real-time trip detection in individuals with lower-limb amputations.
• To assess the performance and generalization capabilities of different detection approaches (thresholds, machine learning, anomaly detection).
• To achieve rapid and reliable perturbation detection for integration into prosthetic leg control systems.

Main Methods:

• Developed three distinct trip detection algorithms: threshold-based, machine learning-based, and anomaly detection-based.
• Validated algorithms using data from 17 healthy subjects and 8 amputees (7 transtibial, 1 transfemoral) on a specialized tripping platform.
• Acquired data capturing gait responses to unanticipated tripping perturbations.

Main Results:

• All developed algorithms demonstrated high performance, with median sensitivities exceeding 80% and specificities above 90%.
• Algorithms trained on healthy subject data maintained high accuracy on amputee datasets, indicating strong generalization.
• Detection times were consistently below 150 ms, with the threshold algorithm achieving the fastest average detection at approximately 40 ms.

Conclusions:

• The developed trip detection algorithms are viable solutions for real-time perturbation detection in prosthetic legs.
• These algorithms offer fast detection times and high specificity, crucial for preventing falls in amputees.
• The study highlights the potential for robust and generalizable trip detection across different user populations.