Brain-machine interface using nonlinear Kalman filters and channel selection

Abstract

This dissertation describes the use of Kalman Filter to decode neural signals, which were recorded using cortical surface potentials, acquired with dense grids of microelectrodes, for brain-computer interfaces (BCIs). This work represents a combination of electronic and biomedical engineering, machine learning and neural science. Kalman filters have been used to decode neural signals and estimate hand kinematics in many studies. However, most prior work assumes a linear system model, an assumption that is violated by neural systems. In this dissertation, I added nonlinearities to the decoding algorithm improving the accuracy of tracking hand movements using neural signal acquired via a 32-channel micro-electrocorticographic (µECoG) grid placed over the arm and hand representations in the motor cortex. Experimental comparisons indicate that a Kalman filter with a fifth order polynomial generative model with cross product relating the hand kinematics signals to the neural signals improved the mean-square tracking performance in the hand movements over a conventional Kalman filter employing a linear system model. While in other works the channel delays were estimated using algorithm based on hill climbing or assuming the uniformity of delay across all the channels. In this work, Particle Swarm of Optimization was applied to better estimate the delays. Also, I was also able to develop a generalized feature selection algorithm and apply to it, to select the most significant channels. As expected this caused a loss in accuracy, but the results of a 16 neural channels system were comparable with the full 32 channel system. This dissertation represents a comprehensive investigation of addition of non linearities, delay estimation and feature selection for Kalman Filter, when used as interface between man and machine.