Unsupervised Spike Sorting Using Local Adaptive Projection and Gaussian Mixture Model

Many researches in neuroscience rely on the analysis of neuronal spike activities recorded under different behavioral conditions, due to the fact that different types of spikes recorded by multi-channel microelectrode arrays may show specific firing in different cognitive tasks. Therefore, accurate and reliable sorting of spikes plays an important role in the related researches in neuroscience. Based on the nonstationarity and local amplitude jump characteristics of spikes, a novel spikes sorting algorithm based on unsupervised local adaptive projection (LAP) feature selection algorithm and Gaussian mixture model (GMM) clustering algorithm (LAGM) was proposed in this work. Firstly, the adaptive similarity matrix and projection matrix were learned by iterative method to select the features of the original spike waveforms using LAP, and the low-dimensional feature set of the spikes was obtained. Then, these low-dimensional features were clustered for spikes sorting based on GMM. Finally, the feasibility and effectiveness of the LAGM was assessed by simulation data from the publish database, and the practicability was validated by the real signals recorded from the hippocamp region of pigeons under the free-foraging with the GMM clustering based on principal component analysis (PCA) dimension reduction and K-means clustering based on LAP feature selection as the algorithms for comparison. Experimental results indicated that the proposed LAGM in this paper shown high accuracy for spike sorting, which provided an effective means for the reliable analysis of spike signals.