Continuous decoding of grasping tasks for a prospective implantable cortical neuroprosthesis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Robotics

Background:

• Invasive cortical neuroprostheses have advanced, enabling control of computer cursors and robotic devices using motor cortex (MI) signals.
• However, predicting different types of grips remains a challenge in neuroprosthetic development.

Purpose of the Study:

• To investigate the continuous decoding of multiple grip types using neural signals from the ventral premotor cortex (area F5).
• To assess the feasibility of using a Support Vector Machines (SVMs) classifier for grip type prediction.

Main Methods:

• Recorded neural signals from motor neurons in area F5 of two monkeys during a reach-to-grasp task.
• Trained a Support Vector Machines (SVMs) classifier to decode four to six different grip types based on recorded neural activity.

Main Results:

• Achieved classification accuracy exceeding 96% for decoding four to six grip types.
• Optimal decoding performance was observed using a time window width of 75-150 milliseconds.

Conclusions:

• The findings demonstrate a high degree of accuracy in decoding grip types from neural signals.
• These results indicate significant potential for developing advanced invasive cortical neural prostheses for controlling reaching and grasping actions.