A New Template Matching Method using Variance Estimation for Spike Sorting.

The analysis of single unit recording data requires a spike sorting method to separate blended neuronal spikes into separate neuron classes. A new template matching method for spike sorting based on shape distributions and a weighted Euclidean metric is proposed. The data is first roughly clustered using a Euclidean distance metric. Then the Levenberg-Marquardt method is used to estimate the variances of the neuron classes using curve fitting on the clustered data. Finally, the weighted Euclidean distance method is applied to minimize errors caused by different variances. This method provides optimized template matching results when the neuron variances are considerably different