Using Lower Limb Wearable Sensors to Identify Gait Modalities: A Machine-Learning-Based Approach

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Real-world gait analysis offers valuable insights for clinical assessments and interventions outside laboratory settings.
• Individual accelerometers provide a feasible method for non-intrusive gait monitoring.

Purpose of the Study:

• To evaluate a simple machine learning method for discriminating between three self-selected cyclical locomotion types using accelerometers.
• To determine the accuracy of gait modality discrimination at different sensor attachment locations (sacrum, thighs, shanks).

Main Methods:

• Thirty-five participants walked at self-selected speeds along a 10m walkway with triaxial accelerometers attached to the sacrum, thighs, and shanks.
• Accelerometer data (3-second intervals) were transformed into 2D Fourier spectra.
• Principal Component Analysis (PCA) and Discriminant Function Analysis (DFA) were used for data reduction, feature selection, and classification.

Main Results:

• The machine learning algorithm achieved high accuracy in discriminating gait modalities across all sensor locations.
• Accuracy rates were 91% at the sacrum, 90% at the shanks, and 87% at the thighs.
• The sacrum location demonstrated the highest accuracy for gait modality discrimination.

Conclusions:

• Simple machine learning algorithms can effectively discriminate between different gait modalities using accelerometer data from common body locations.
• The sacrum offers the most accurate sensor placement for this type of real-world gait analysis.
• Future research should optimize data processing for sensor locations beneficial for specific applications, such as shank placement for prosthetic devices.