Advantage of support vector machine for neural spike train decoding under spike sorting errors

Abstract

Decoding of kinematic variables from neuronal spike trains is important for neuroprosthetic devices. The spike trains from single units must be extracted from extracellular neural signals and thus spike detection and sorting procedure is essential. Since the spike detection and sorting procedure may yield considerable errors, decoding algorithm should be robust against spike train errors. Here we showed that the spike train decoding algorithms employing a nonlinear mapping, especially support vector machine (SVM), may be more advantageous contrary to conventional belief that linear filter is sufficient. The advantage became more conspicuous with erroneous spike trains. Using the SVM, satisfactory performance could be obtained much more easily, compared to the case of using multilayer perceptron, which was employed for previous studies. The results suggests the possibility of neuroprosthetic device with a low-quality spike sorting preprocessor.