Design, development, and evaluation of a local sensor-based gait phase recognition system using a logistic model...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Microprocessor-controlled stance-control knee-ankle-foot orthoses (M-SCKAFO) rely on control systems for functionality.
• Accurate gait phase recognition (GPR) is crucial for M-SCKAFO to prevent knee-collapse and falls.
• Localizing sensors to the thigh and knee can enable more flexible orthosis designs.

Purpose of the Study:

• To investigate the effectiveness of machine learning using local thigh and knee sensor signals for gait phase recognition.
• To determine if gait phase transition criteria improve GPR performance across various walking conditions.
• To test the hypothesis that local sensor data can effectively distinguish gait phases.

Main Methods:

• A logistic model decision tree (LMT) classifier was trained on gait data (knee flexion angle, angular velocity, acceleration).
• Features were extracted from sliding windows for participants walking on diverse surfaces and speeds.
• A Transition Sequence Verification and Correction (TSVC) algorithm was applied to refine GPR accuracy.

Main Results:

• The LMT-based GPR model achieved high accuracy (98.38% training, 90.60% validation).
• Applying the TSVC algorithm improved validation accuracy to 98.61%.
• Performance metrics indicate strong feasibility for real-time orthosis control.

Conclusions:

• A novel machine learning GPR model using local sensors is viable for M-SCKAFO control.
• The GPR model, enhanced by TSVC, demonstrates high accuracy and generalizability.
• This approach simplifies sensor systems, allowing for customizable and modular orthosis designs.