A feature-based algorithm for spike sorting involving intelligent feature-weighting mechanism

Spike sorting of neural data from multiple electrodes is a difficult problem that depends heavily on inputs from human experts. It is an important processing step in the study of various brain functions and to detect various neural disorders based on the activity of neurons. Here, we propose a novel, unsupervised, feature-based spike sorting method based on the K-means clustering algorithm to distinguish these spikes. It involves weighing the various features of the neural data based on their information content as well as the eigenvalues of their projections on the lower-dimensional space and clustering them in the absence of ground truth. We illustrate the method on simulated data and real data recorded from retinal degeneration (rd) mice. We also compared our method against previously reported algorithms such as principal component analysis (PCA) based spike sorting and the results found are very encouraging for determining the activity of each neuron and early detection of various neural disorders including blindness (Retinitis Pigmentosa).