Automatic Classification of Neural Data

In this thesis we present a new solution for an automatic classification of the single-neuron activity. The study of the computational role of individual neurons underlying different cognitive process is a gold standard in Neuroscience. This type of analysis is done first, by recording the extracellular spikes of the neurons near the tip of a microelectrode and second, by isolating the spikes of the recorded cells based on the similarity of their shapes using a method called spike sorting. In recent years, important advances in microelectrode technology allow us now to perform massive parallel recordings using a high number of channels with the possibility to study the activity of large ensembles of neurons at a time. However, this fascinating opportunity introduces at the same time a challenge for the efficient and fast analysis of this data. In this research work, we address this problem by developing a new implementation for unsupervised spike sorting that improves the performance of a widely-used spike sorting algorithm, increasing the number of automatically identified neurons. Moreover, we developed a new testing platform which generates simulations of extracellular recordings including challenging conditions such as realistic noise, multi-unit activity -spikes of distant neurons impossible to be identified as single units- or the presence of neurons with low firing rates. In summary, the results presented here provide contributions to the development of automated and efficient quantitative frameworks for the analysis of multiple-channel recordings that help us to understand single-neuron population codes.