Figure 2.

Illustration of the procedure of the proposed automatic spike sorting algorithm. First, the raw signal is fed into a common detection procedure. As an example, first, the signal is filtered, then, an initial detection is performed by applying a simple thresholding procedure. Thereafter, the proposed adaptive detection procedure takes the initially detected spikes and performs a set of procedures to improve the detection results and make them ready for feature extraction. In adaptive detection, first, the noise which is detected as waveforms are removed using the proposed statistical filtering. Then the proposed multi-point alignment seeks to align the waveforms based on the histogram of the extremums. Then, an adaptive number of principal components are considered as features. Finally, the samples are fed to the clustering method based on the mixture of skew-t distributions.