Automotic Recognition of Sleep Spindles Based on Two-Stage Classifier with Artificial Neural Networks and Support Vector Machines

Abstract

Sleep spindles are one of the most important transient waveforms found in the sleep EEG signal. Here, we introduce a two-stage procedure based on artificial neural networks for the automatic recognition of sleep spindles (SS) in a 19-channel electroencephalographic signal. In the first stage, a pre-processing perception is used for enhancing overall detection and also reducing computation time. In the second stage, the selected Sleep spindles (SS), classified with neural network post-classifier. Classifying tools in post-processing procedure were MLP and RBSVM that their operations are compared in the last section of the report. Visual inspection of 19-channel EEG from six subjects by one expert in this theme, showed that RBSVM operation is better than MLP with BP (Back propagation) training, that SVM provided 91.4%  average sensitivity and 3.85% average false detection rate.