Abstract
The long-term stability and low frequency composition of local field potentials (LFPs) offer significant advantages for the development of robust, low-power, implantable neuroprostheses. However, our poor understanding of the information contained within these signals currently hinders their use in applications beyond supervised (biomimetic) decoding of movement kinematics. Our research aims to understand how LFPs relate to activity within cortical networks, and to improve methods to extracting features from multichannel LFPs for brain interfacing. In the first part of my talk I will describe a method to decode neuronal firing rates from LFPs that exploits a phenomenon we call the spike-related slow potential (Hall et al. Nature Communications 2014). This technique enables surprisingly accurate estimation and biofeedback control of single neuron firing rates long after the extracellular spike recordings have been lost. However, a disadvantage of this method is the requirement for initial spike recordings used to calibrate the decoder. In the second part of my talk, I will discuss a new approach to LFP decoding based on low-frequency cycles that reflect intrinsic constraints on network dynamics (Hall et al. Neuron 2014). The areal velocity of LFP cycles contains multidimensional information about movement kinematics as well as the activity of local neurons. Areal velocity is simple to compute, robust to noise, stable over long periods and can be readily brought under biofeedback control to enable multidimensional cursor control. We suggest that this may provide a promising new approach to unsupervised feature extraction for LFP-based neuroprostheses.