BME Seminar Series: Moving Beyond Linear State-Space Models to Advance Brain-Machine Interfaces
Thursday, February 4, 2016
River Campus | Robert B. Goergen Hall | Sloan Auditorium (Room 101)
Speaker: Adam Rouse, MD, PhD, Department of Neuroscience, University of Rochester
The field of motor brain-machine interfaces (BMIs) has advanced dramatically. Our ability to accurately decode neural activity to directly control a cursor, robotic arm, or the patient’s own muscles continues to improve. However, this control remains robotic and limited compared with natural human performance. Most BMI decoding relies on each neuron having a fixed and linear relationship to a given set of degrees of freedom. In experimental results from a reach-to-grasp task I will describe the sequential phases of movement observed with EMG, kinematic, and single-unit neurophysiologic recordings. Additionally, I will show the broad tuning throughout the entire upper forelimb region of primary motor cortex to both reach location and grasp object type and how it transitions between phases of the movement.
I will also demonstrate why this sequential, selective tuning can serve as an important principle for BMI design. By using active dimension selection and 4 ethologically relevant dimensions of control, I will show how a simple 16 single unit BMI can efficiently control a virtual hand to achieve 8 different postures with 95% accuracy with average movement times of ~1 second.