A support vector approach to online brain control

This paper introduces a new approach to direct motor cortical brain machine interfaces between rats and a directional decision task. This approach does not rely on a model of neural activity like preferred direction approaches. It builds a functional mapping between the neural firing patterns and the control command using a kernel based method, namely the support vector machine. Therefore this approach seeks to classify the spatial and temporal neural firing activities without making assumptions on how the neural signals produce movement in intact individuals. Offline analysis offers critical evaluation of parameters used in decoding the control command from the cortical signal. Further these results show the effect of increasing the number of neural signals used on the accuracy of the control commands generated. The method was then implemented in a real-time, closed-loop situation where four animals, after a brief paddle pressing phase for calibration, used signals from their motor cortex to actuate the task achieving an accuracy of 77.7% (71.7% 85.8%). These results hold hope for a possibly robust and realistic interface capable of adding quality to the life of paralyzed people.