Prediction of hand path and reach location from neuronal activity in dorsal premotor cortical area for neuroprosthetic control

This project is part of an ongoing effort to develop advanced neuroprosthetic control by extracting decision commands of future hand path from cortical signals. A novel experiment with presentation of a virtual obstacle in an instructed-delay paradigm was designed to study the effect of different types of intended trajectory on cortical activity in dorsal premotor (PMd) area.; The cortical activity in 687 PMd neurons was studied in two rhesus monkeys trained to perform a 3-D delayed point-to-point reaching task based on visual cues of target location, conditional appearance of an obstacle and finally the go signal. One of major findings is that among 687 set-related neurons, 165 (24%) neurons were modulated by obstacle presence or absence (2 nd visual cue) in their onset time and another 206 (30%) neurons showed modulation by obstacle cue in their overall discharge level during set period. In addition, 110 (16%) and 97 (14%) neurons were modulated by target location (1st visual cue) only and both cues.; A nonlinear back-propagation artificial neural network (ANN) model was developed to predict different types of intended hand path from single neural activity during movement preparation. Three temporal parameters, bin size, window duration and latency, integrating spikes within time windows relative to trial events, were assessed by evaluating their prediction accuracy of the target learning algorithm. The optimal value of each parameter was found to be correlated to the characteristics of neural activity. Besides, the prediction accuracy was improved using two and three sets of simultaneously recorded neurons compared to single neuron.; In addition to studying discharge patterns in successful trials, unsuccessful trials were investigated as well. One major type of unsuccessful trials is caused by hitting the virtual obstacle, termed as "obstacle hit". In "obstacle straight hit" trials, the cortical activity in neurons modulated by obstacle presence was similar to the pattern from successful trials with straight hand path. A quantitative analysis also indicated similar prediction result between "obstacle hit" trials and successful trials with straight hand path. Therefore decision of forthcoming handpath was able to be extracted from cortical activity in PMd. Both successful and unsuccessful movements indicated that the cell activity reflected planning of different type of future hand path.