Interpreting neural activity through linear and nonlinear models for brain machine interfaces

Brain machine interface (BMI) design can be achieved by training linear and nonlinear models with simultaneously recorded cortical neural activity and behavior (typically the hand position of a primate). We propose the use of optimized BMI models for analyzing neural activity to assess the role of individual neurons and cortical areas in generating the performed movement. Two models (linear-feedforward and nonlinear-feedback) are trained to predict the hand position of a primate from neural recordings in a reaching task. Qualitative and quantitative investigation of the effect of neurons and their corresponding cortical areas through both models yields conclusions consistent with neurophysiologic knowledge. In addition, this analysis revealed the role of these areas and the importance of the neurons in terms of BMI design.