Bayesian clustering and tracking of neuronal signals for autonomous neural interfaces

This paper introduces a new, unsupervised method for sorting and tracking the non-stationary spike signals of individual neurons in multi-unit extracellular recordings. While this method may be applied to a variety of problems that arise in the field of neural interfaces, its development is motivated by a new class of autonomous neural recording devices. The core of the proposed strategy relies upon an extension of a traditional expectation-maximization (EM) mixture model optimization to incorporate clustering results from the preceding recording interval in a Bayesian manner. Explicit filtering equations for the case of a Gaussian mixture are derived. Techniques using prior data to seed the EM iterations and to select the appropriate model class are also developed. As a natural byproduct of the sorting method, current and prior signal clusters can be matched over time in order to track persisting neurons. Applications of this signal classification method to recordings from macaque parietal cortex show that it provides significantly more consistent clustering and tracking results than traditional methods.