Using Deep Learning to Predict Minimum Foot-Ground Clearance Event from Toe-Off Kinematics.

Area of Science:

• Biomechanics
• Machine Learning
• Wearable Technology

Background:

• Efficient locomotion is vital for health and independence, but falls, particularly tripping, pose significant risks for older adults and post-stroke individuals.
• Minimum Foot Clearance (MFC) during the swing phase of gait is the primary biomechanical factor determining tripping probability.
• Current methods for measuring MFC rely on laboratory-based 3D motion capture, lacking real-world applicability for intervention.

Purpose of the Study:

• To develop and validate Machine Learning (ML) algorithms for predicting the timing of Minimum Foot Clearance (MFC) using preceding gait events.
• To enable real-time identification of MFC for the development of assistive technologies to prevent tripping.

Main Methods:

• Utilized foot trajectory data from 13 young adults captured via an Optotrak 3D motion capture system.
• Developed a Deep Learning model employing a Recurrent Neural Network with Long Short-Term Memory (LSTM) architecture.
• Optimized the model using Huber loss functions to minimize MFC timing prediction error.

Main Results:

• Successfully predicted MFC timing from toe-off characteristics with a mean absolute error of 0.07 seconds.
• Demonstrated the feasibility of using ML algorithms to identify critical MFC timing outside laboratory settings.

Conclusions:

• The developed ML algorithms offer a pathway for real-time MFC prediction, crucial for actuating wearable devices to increase foot clearance and reduce tripping falls.
• Further refinement with population-specific data is necessary, but this approach holds significant potential for gait-impaired populations and advanced exoskeleton technologies.