Classification of motor commands using a modified self-organising feature map

Area of Science:

• Biomedical Engineering
• Neuroscience
• Artificial Intelligence

Background:

• Advanced prosthetics require sophisticated control systems.
• Existing control systems face challenges in real-time adaptation and accuracy.
• Biomimetic approaches offer potential for improved prosthetic functionality.

Purpose of the Study:

• To propose and investigate a novel control system for advanced prostheses.
• To evaluate the system's performance in both animal models and human subjects.
• To compare the efficacy of different artificial neural network (ANN) algorithms for prosthetic control.

Main Methods:

• Utilized a rat spinal withdrawal reflex system as a model for prosthetic control.
• Validated the control system in human subjects, including one with traumatic hand amputation.
• Employed various ANN algorithms, including a modified Self-Organising Feature Map (SOFM) combining Kohonen network and conscience mechanism (KNC), for classifying electromyography (EMG) signals.
• Compared KNC performance against reference networks like multi-layer perceptrons.

Main Results:

• The modified SOFM (KNC) algorithm outperformed reference networks in training time, memory consumption, and ease of parameter optimization.
• KNC achieved high classification accuracy for five movements in the animal model.
• KNC demonstrated high accuracy in classifying seven distinct movements in human subjects.

Conclusions:

• The proposed KNC-based control system offers a superior, efficient, and accurate method for advanced prosthetic control.
• This approach shows significant promise for enhancing prosthetic functionality in both simulated and real-world applications.
• The study validates the effectiveness of advanced ANNs in decoding complex biological movement signals for prosthetic interfaces.