Real-Time Spike Sorting for Multi-Electrode Arrays with Online Independent Component Analysis

When recording neural activity from extracellular electrodes, spike sorting is needed to separate the activity of different neurons. Most of the spike sorting packages are offline and use all the available data to distinguish the activity of single neurons from the recordings. However, when performing an experiment, it is helpful to monitor the activity of recorded neurons. Real-time spike sorting is not a trivial problem and previous approaches usually require the user to manually separate units or an initial offline calibration phase, which can be deleterious in case of non-stationarity of the recordings. In this contribution, we adapted the Online Recursive Independent Component Analysis (ORICA) algorithm for real-time spike sorting of high-density Multi-Electrode Array (MEA) data. Our approach, with its recursive implementation and dimensionality reduction, has the potential to cope with non-stationarity of the signals (e.g. caused by electrode drift) and it yields a better representation of the neural data, which facilitates spike detection and improves the detection accuracy.