NeuroNella: A Robust Unsupervised Algorithm for Identification of Neural Activity from Multielectrode Arrays

Abstract

We introduce NeuroNella, an automated algorithm developed for the identification of neuronal activity from multichannel electrode arrays. In evaluations conducted on recordings from implanted probes in the nervous system of rodents and primates, the algorithm demonstrated remarkable accuracy, showcasing an error rate of less than 1% compared to ground-truth patch clamp signals. Notably, the proposed algorithm handles large datasets efficiently without the necessity of a GPU system. The results highlighted the algorithms efficacy in detecting sources in a wide amplitude range and its adaptability in accommodating minor probe shifts. Moreover, the high robustness exhibited by the algorithm in decomposing recordings lasting up to 30 minutes underscores its potential for enabling longitudinal studies and prolonged recording sessions, thus opening new avenues for future brain/machine interface applications.