Vision-based segmentation of continuous mechanomyographic grasping sequences for training multifunction prostheses.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Acquiring long, continuous mechanomyographic (MMG) signals is crucial for prosthetic control design.
• Separating individual muscle contractions from continuous MMG data streams presents a significant challenge.

Purpose of the Study:

• To develop and validate an automatic, vision-based segmentation method for continuously recorded MMG data streams.
• To improve the efficiency of data processing for MMG signal classifiers used in prosthetic control.

Main Methods:

• Synchronized acquisition of MMG and transverse plane video data during functional grip sequences.
• An automatic system was developed to track hand movements, recognize grips, and detect grip transition times.
• The system also identified and accounted for extraneous hand movements.

Main Results:

• The vision-based system achieved high accuracy in recognizing grips (97.8 +/- 4% for two grips, 73 +/- 20% for seven grips).
• Contraction initiation and termination times were accurately identified, with close agreement (1.3 +/- 1 frames) to manual segmentation.
• The method effectively segments continuous MMG data streams into individual contractions.

Conclusions:

• Automatic, vision-based segmentation is a viable and accurate method for processing continuous MMG data.
• This technique enhances the development of MMG signal classifiers for advanced prosthetic control systems.
• The proposed method offers a more efficient alternative to manual segmentation of MMG data.