Tetrode Spike Detection Method Based on Quaternion Principle Component Feature Extraction

Abstract

Abstract-Multi-unit recording of tetrode has been used in spike detections for many years, but traditional feature extraction methods of spikes sorting based on analyzing correlations of single channel data don't consider relations of each channel data. A new feature extraction method based on quaternion principle component analysis (QPCA) algorithm used for sorting the tetrode spikes trains is presented. A quaternion vector formed of fourdimensional numbers can replace a set of tetrode spikes data. Firstly, the four channels of spikes were achieved using dual threshold detection method. We used the modulus values of the vectors grouped by 4 channels corresponding data to constitute the new features vectors of spikes through QPCA algorithm. The features vectors contain correlations of 4 channels. Our method fully fuses the information of tetrode data. Thus it has higher clustering accuracy than traditional feature extraction methods.