A switching Kalman filter model for the motor cortical coding of hand motion

We present a switching Kalman filter model (SKFM) for the real-time inference of hand kinematics from a population of motor cortical neurons. First we model the probability of the firing rates of the population at a particular time instant as a Gaussian mixture where the mean of each Gaussian is some linear function of the hand kinematics. This mixture contains a "hidden state", or weight, that assigns a probability to each linear, Gaussian, term in the mixture. We then model the evolution of this hidden state over time as a Markov chain. The expectation-maximization (EM) algorithm is used to fit this mixture model to training data that consists of measured hand kinematics (position, velocity, acceleration) and the firing rates of 42 units recorded with a chronically implanted multi-electrode array. Decoding of neural data from a separate test set is achieved using the switching Kaiman filter (SKF) algorithm. Quantitative results show that the SKFM outperforms the traditional linear Gaussian model in the decoding of hand movement. These results suggest that the SKFM provides a real-time decoding algorithm that may be appropriate for neural prosthesis applications.