PMd Representation And Decoding Of Monkey Reach Plan And Execution During Obstacle Avoidance Task

Brain machine interfaces build a direct pathway between brain cortex and external devices,which brings hope for the movement rehabilitation of paralytics.The reaching movement of upper limb is one of the mostly utilized functions in daily life.In practice,there exists obstacles between upper limb and target.It is significant for paralytics in both physiology and psychology to restore the movement ability of upper limb in obstacle avoidance task.Neuroscientists found that the reaching movement is mainly controlled by the dorsal-medial pathway,in which the dorsal premotor cortex(PMd)is closely related to both movement planning and execution.However,the tuning properties of PMd neurons during obstacle avoidance task is not clear;the trajectory decoding of obstacle avoidance task challenges the common decoding models;it is a challenge to achieve the online real-time obstacle avoidance task decoding and robot arm controlling.In this paper,we focused on the following studies:1)representation of PMd during obstacle avoidance planning;2)design of neural decoding model integrating the planning information during movement execution;3)online real-time decoding and controlling of robot arm doing obstacle avoidance task.In this study,we trained two adult male rhesus monkeys doing the delayed obstacle avoidance task.The neural signal of PMd and movement trajectory were recorded simultaneously.We studied the representation of PMd during planning period in a view of single neuron and neuron populations.The result shows that PMd has significant representation of initial cursor position,target position,the relative direction from initial cursor to target and obstacle information.Based on the tuning properties of PMd,the mixture of trajectory models integrating target direction and movement selection during planning were designed.The trajectory estimation performance was promoted by 15%compared with model without prior information.The asynchronous online real-time control system was designed based on the proposed decoding models in two steps.Support vector machine(SVM)was utilized to classify the experiment states with different operations:decoding model did not work during the rest state;the target direction and movement selection were extracted by naive Bayes classifier during the planning state;decoding with planning information and controlling the robot arm doing obstacle avoidance task during movement state.The innovations of this study are:1)the movement parameters representation of PMd was carried out on single neuron and neuron population and the results showed that PMd has signifi-cant representation to cursor position,target position,direction from cursor to target and obstacle avoidance selection;2)designing and implementing the decoding algorithm with integration of planning information and improving the trajectory prediction performance and trial successful rate;3)achieving the real-time neural decoding of obstacle avoidance task and movement trajectory reestablishment by robot arm.The works above lay a solid foundation for the practical efficient brain machine interfaces.