Combining two open source tools for neural computation (BioPatRec and Netlab) improves movement classification for...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Prosthetics

Background:

• Controlling advanced myoelectric prostheses with multiple joints is complex.
• Pattern recognition using multi-electrode arrays enables simultaneous control of prosthetic joints.
• Lack of standardized datasets hinders comparison of previous pattern recognition studies.

Purpose of the Study:

• To compare the performance of different pattern recognition models for myoelectric prosthesis control.
• To identify the optimal algorithm and network model for improved prosthetic function.
• To facilitate standardized evaluation by analyzing open-access datasets.

Main Methods:

• Utilized BioPatRec, an open-source Matlab platform, for feature extraction and motion classification.
• Applied artificial neural networks and linear models to myoelectric signals.
• Compared Netlab and BioPatRec pattern recognition models using scaled conjugate training algorithms.

Main Results:

• Both linear and artificial neural network models showed Netlab's scaled conjugate training algorithm achieved higher accuracies than BioPatRec.
• Netlab's implementation demonstrated superior performance in myoelectric signal classification.
• Classification accuracy was a key evaluation criterion, with Netlab showing significant improvements.

Conclusions:

• Integrating Netlab training algorithms into the BioPatRec environment is recommended for optimal movement classification.
• This integration aims to shorten prosthesis training time and increase control reliability.
• Netlab has been incorporated into BioPatRec version 4.0 to enhance prosthetic control capabilities.