Assessing Spike Sorting contribution to neural implant reliability

Recording up to hundreds of neural electrodes simultaneously is nowadays possible thanks to many commercially available acquisition systems. However the reliability of these data for subsequent encoding or decoding analyses strictly depends on the capability of specific algorithms to identify single neuron contribution from multi-unit activity acquired by each channel. This is an ill-posed problem, which is called Spike Sorting (SS). So far, several algorithms have been proposed, each based on specific features of the recorded electrophysiological data. The spike sorting is the very first phase of the chain analysis of neural spiking activity: an error at this stage would severely propagate further. In this study we investigate how reliable the results of SS algorithms are, by using simulated neural data at different recording conditions. Our findings support the idea that fallibility in attaining stable single-unit neural signals across different recording sessions is not only due to technical features of the neural recording implant, but may significantly depend on propagated error from the SS stage.