Adaptive myoelectric pattern recognition for arm movement in different positions using advanced online sequential...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Myoelectric pattern recognition (MPR) systems show reduced performance with varying limb positions.
• Traditional training methods to account for limb position variations are cumbersome and user-unfriendly.
• Unpredictable environmental scenarios pose challenges for existing MPR systems.

Purpose of the Study:

• To propose a novel adaptive MPR system using advance online sequential extreme learning (AOS-ELM).
• To enhance the classification accuracy of hand movements across diverse limb positions.
• To improve the real-time adaptability and user comfort of MPR systems.

Main Methods:

• Development of AOS-ELM, an enhanced version of OS-ELM with entropy-based adaptation validity verification.
• Classification of hand movements into five distinct positions.
• Evaluation of the system's performance across multiple limb positions and subjects.

Main Results:

• The proposed adaptive MPR achieved 95.42% accuracy for eight classes from eleven subjects using single-position data.
• After learning data from all positions, the system achieved 86.13% accuracy.
• The AOS-ELM based MPR outperformed the original OS-ELM but was less accurate than batch classifiers.

Conclusions:

• The AOS-ELM based adaptive MPR system demonstrates improved performance and adaptability compared to standard OS-ELM.
• While not surpassing batch classifiers in accuracy, the adaptation mechanism is highly suitable for real-time applications.
• The proposed system offers a promising solution for robust hand movement classification in dynamic environments.