Automatic spike sorting by unsupervised clustering with diffusion maps and silhouettes

Knowledge of the activity of single neurons is crucial for understanding neural functions. Therefore the process of attributing every single spike to a particular neuron, called spike sorting, is particularly important in electrophysiological data analysis. This task however is greatly complicated because of numerous factors. Bursts or fast changes in ion channel activation or deactivation can cause a large variability of spike waveforms. Another considerable source of uncertainties results from noise caused by firing of nearby neurons. Movement of electrodes and external electrical noise from the environment also hamper the spike sorting. This paper introduces an integrated approach of diffusion maps (DM), silhouette statistics, and k-means clustering methods for spike sorting. DM is employed to extract spike features that are highly capable of discriminating different spike shapes. The combination of k-means and silhouette statistics provides an automatic unsupervised clustering, which takes features extracted by DM as inputs. Experimental results demonstrate the noticeable superiority of the features extracted by DM compared to those selected by wavelet transformation (WT). Accordingly, the proposed integrated method significantly dominates the popular existing combination of WT and superparamagnetic clustering regarding spike sorting accuracy.