Development of a Start-Stop Signal for a Directional BMI.

While BMI systems abound, little care has been exercised over practical considerations in the day to day use of such systems. This paper proposes to learn a Start-Stop switch to augment a directional BMI. Taken together, the hope is that a BMI could be constructed that would be able to signal the appropriate directional intent when called upon and be virtually silent when not needed. Using data from rats utilizing a simple directional BMI, an attempt is made to test several possible methods for integrating a Start-Stop decision. Three methods, a 3-class SVM, directional classifier with probabilistic output, and a directional classifier with Start-Stop modulated probabilistic output are constructed and compared. Results show that the directional classifier with Start-Stop performs well with a significant reduction in signaling outside the task period