Feature extraction of somatosensory evoked potentials based on ICA for classification of external tactile stimuli in rat

Four neural signals are recorded by without stimulation, by stimulation using a toothbrush, pen shaft and needle under an anesthetized rat. First, spectral subtraction is used to reduce noise and the nonlinear energy operator is adopted to detect spikes. Then, independent component analysis is performed with dynamic dimension increase to extract the features and form a feature vector. Finally, k-means is employed to group the feature vector into different clusters. These four various evoked potentials are separated into respective cluster according to differential percentage of 100%, 67%, 43%, and 73% individually. The information of monitoring subsystem is applied to assist us in proving of experimental results. The presented methods are successfully utilized to extract the features from various evoked potentials and distinguish the stimulants from different sensory signals.