Swarm-wavelet based extreme learning machine for finger movement classification on transradial amputees

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Prosthetics

Background:

• Transradial amputees face challenges in controlling myoelectric prostheses with limited surface electromyography (EMG) channels.
• Developing accurate pattern recognition systems for prosthetic control is crucial for improving user dexterity and function.

Purpose of the Study:

• To propose and evaluate a novel pattern recognition system for classifying imagined finger motions in transradial amputees using minimal EMG channels.
• To optimize an extreme learning machine (ELM) classifier with a wavelet-mutated particle swarm optimization (PSO) algorithm.

Main Methods:

• A pattern recognition system was developed using an extreme learning machine (ELM).
• Particle swarm optimization (PSO) was employed to optimize the ELM, with the PSO algorithm mutated by a wavelet function to prevent local minima.
• The system was trained and tested on five transradial amputees, classifying eleven imagined finger motions using only two EMG channels.

Main Results:

• The proposed wavelet-PSO optimized ELM system achieved an average classification accuracy of approximately 94% across five amputees.
• The performance of the wavelet-PSO method was superior to both grid-search and standard PSO optimization techniques.
• The system demonstrated high accuracy in classifying multiple imagined finger movements with a reduced number of EMG channels.

Conclusions:

• The wavelet-PSO-ELM system offers a highly accurate and efficient solution for myoelectric control in transradial amputees.
• Utilizing a minimal number of EMG channels with advanced optimization techniques can significantly enhance prosthetic function.
• This approach holds promise for developing more intuitive and responsive prosthetic devices.