Utilizing Inertial Measurement Units for Detecting Dynamic Stability Variations in a Multi-Condition Gait Experiment

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Gait assessment is crucial for evaluating mobility and stability.
• Current methods often require controlled laboratory settings.
• Wearable sensors offer potential for unobtrusive, real-world gait analysis.

Purpose of the Study:

• To develop and validate a wearable gait assessment method using inertial measurement units (IMUs).
• To estimate the Margin of Stability (MoS) from IMU acceleration data using a convolutional neural network (CNN).
• To investigate the relationship between MoS and other gait stability indices under various conditions.

Main Methods:

• Utilized five IMUs placed on different body parts to collect acceleration time-series data.
• Developed a CNN model to estimate MoS from IMU data.
• Simulated diverse gait conditions by varying treadmill speed and inducing asymmetry with a knee-ankle-foot orthosis.
• Examined correlations between MoS, Lyapunov exponent, and multi-site time-series (MSTS) index.

Main Results:

• The CNN model achieved over 90% accuracy in classifying MoS in both forward and lateral directions.
• A weak correlation was observed between MoS and the Lyapunov exponent or MSTS index.
• The findings suggest these indices may reflect distinct aspects of gait stability.

Conclusions:

• A wearable IMU-based method using CNNs can accurately estimate MoS for gait assessment.
• This approach shows potential for evaluating gait stability in daily environments.
• Further research is needed to understand the complementary information provided by different stability indices.