Interpreting Spatial and Temporal Neural Activity Through a Recurrent Neural Network Brain–Machine Interface

J.C. Sanchez, D. Erdogmus, M.A.L. Nicolelis, J. Wessberg, J.C. Principe
2005 IEEE transactions on neural systems and rehabilitation engineering  
We propose the use of optimized brain-machine interface (BMI) models for interpreting the spatial and temporal neural activity generated in motor tasks. In this study, a nonlinear dynamical neural network is trained to predict the hand position of primates from neural recordings in a reaching task paradigm. We first develop a method to reveal the role attributed by the model to the sampled motor, premotor, and parietal cortices in generating hand movements. Next, using the trained model
more » ... ained model weights, we derive a temporal sensitivity measure to asses how the model utilized the sampled cortices and neurons in real-time during BMI testing. Index Terms-Analysis of neural activity, brain-machine interface (BMI), motor systems, nonlinear models, recurrent neural network, spatio-temporal. I. INTRODUCTION M ANY brain-machine interface (BMI) researchers have demonstrated the feasibility of using adaptive inputoutput models for reconstructing hand trajectories [1]-[8]. All of the proposed models in the literature demonstrated the ability to encode and store the fundamental timing relationships between neural inputs and hand trajectory [9]- [12] . To achieve the mapping, model parameters (weights) are adjusted to minimize the difference between the model output and hand movements using a statistical criterion such as mean-square error. A natural next step is to analyze how the trained models extracted the spatio-temporal trends in the neural recordings. By analyzing the model parameters in a signal-processing context, we can hypothesize relationships between neurons, cortices, motor systems, and behavior. This type of analysis exploits the fact that the trained model embody precise functional relationships between its inputs (the neurons, cortices), and their outputs which are a proxy for the observed behavior. Moreover,
doi:10.1109/tnsre.2005.847382 pmid:16003902 fatcat:3mixj3morrf27jworchk7o4yca