Walking Mode-depending Improvements of Locomotion Detection through Rejection Based Post-Processing.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Prosthetics

Background:

• Intuitive prosthetic control is hindered by limited user intent and environmental data.
• Current machine learning methods using muscle and motion sensors lack the reliability needed for leg prostheses.
• Accurate locomotion mode detection is crucial for effective prosthetic limb function.

Purpose of the Study:

• To improve the accuracy of locomotion mode detection in prosthetic control.
• To enhance classification by incorporating postprocessing after Linear Discriminant Analysis (LDA).
• To refine prosthetic control by addressing uncertainties in locomotion classification.

Main Methods:

• Collected locomotion data from 15 able-bodied participants using electromyography (EMG), inertial measurement units (IMUs), and insole pressure sensors.
• Classified locomotion modes (level walking, stair/ramp ambulation) using Linear Discriminant Analysis (LDA).
• Implemented a threshold-based rejection postprocessing method to eliminate uncertain classifications.

Main Results:

• The postprocessing rejection threshold significantly improved overall locomotion detection accuracy.
• Transition locomotion detection showed more substantial improvement compared to steady-state locomotion.
• Biomechanically similar transitions (e.g., ramp to level walking) improved more than dissimilar ones (e.g., stair to level walking).

Conclusions:

• Postprocessing methods can significantly enhance the accuracy of prosthetic locomotion mode detection.
• Refining uncertain classifications through additional data and postprocessing is vital for prosthetic users.
• This approach offers a pathway to more intuitive and reliable prosthetic control.