Unsupervised spike sorting for multielectrode arrays based on spike shape features and location methods

Microelectrode arrays (MEAs) enable simultaneous measurement of spike trains from numerous neurons, owing to advancements in microfabrication technology. These probes are highly valuable for comprehending the intricate dynamics of neuronal networks. Spike sorting is a pivotal step in comprehensively analyzing the activity of neuronal networks from extracellular recordings. However, the accuracy of spike sorting is relatively low due to the dense sampling of spikes in MEAs. Here, we propose an unsupervised pipeline named UMAP-COM method, which utilizes combined features to address this problem. These combined features comprise dominant spike shape features extracted by the uniform manifold approximation and projection (UMAP), as well as spike locations estimated by the center of mass (COM). We validate the UMAP-COM method on publicly available datasets from different kinds of probes, demonstrating that it is more accurate than other spike sorting methods. Furthermore, we conduct separate evaluations of spike shape feature extraction methods and spike localization methods. In this comparison, UMAP emerges as the superior feature extraction method, demonstrating its effectiveness in accurately representing spike shapes. Additionally, we find that the COM method outperforms other spike localization methods, highlighting its ability to enhance the accuracy of spike sorting.