Functional classification of neurons in mouse hippocampus based on spike waveforms in extracellular recordings

Neurons are functionally classified into inhibitory and excitatory categories based on the influence they have on the firing rates of their postsynaptic neurons after being stimulated. Although assessing the firing rates of postsynaptic neurons is the main way of this categorization, it is very hard in real cases. Due to the lack of a labelled dataset with inhibitory and excitatory neurons, past studies have been conducted to investigate the feasibility of this categorization based on clustering some features of the spike waveforms and evaluating the results by physiological evidence. However, there is still the lack of a classification study in order to do this categorization by using features of spike waveforms and different classifiers. This is what we addressed in this paper based on a recent labeled dataset of mouse hippocampus neurons. We extracted nine different features from neuron spikes. Then we investigated the significance of difference of each feature between inhibitory and excitatory groups using Wilcoxon rank-sum-test and also evaluated the effectiveness of all possible feature subsets for classification using KNN, LDA, and SVM classifiers. The highest average classification accuracy was %96.96 obtained by using SVM with RBF kernel and five features. However, KNN yielded %96.08 average accuracy by using just one feature which was Peak amplitude asymmetry. In addition, Peak amplitude asymmetry, Peak-to-trough ratio, and Duration between peaks selected more in the optimum feature subsets using different classifiers. Generally, we concluded the features obtained from waveform spikes and simple common classifiers can effectively classify neurons into inhibitory and excitatory categories.