An adaptive neuro-fuzzy inference system for sleep spindle detection

In this paper, an adaptive neuro-fuzzy inference system (ANFIS) for sleep spindle detection was developed. Two input variables including teager energy operator (TEO) and sigma index analyses of the EEG signals were extracted. 1180 training samples (0.5 s) of 15 subjects were used to ANFIS training, include 397 spindle and 783 non-spindle waveform. Then the 1519 epochs (30s) of other 15 subjects were used to evaluate the performance of ANFIS. The overall sensitivity and specificity of the ANFIS are 94.09% and 96.76%, respectively. Although the overall false positive rate is 38.58%, spindle and non-spindle successful detection rate could almost reach 90% for each subject. This method can integrate with various PSG systems for sleep monitoring in cognitive enhancements or sleep efficiency.