Detection of cortical arousals in sleep EEG

Abstract

Introduction: Cortical arousals (CA) are a transient part of the sleep-wake system that may play an important role in characterizing sleep fragmentation and disorders, such as obstructive sleep apnea. The American Association of Sleep Medicine (AASM) (1992) rules describe electroencephaolgram (EEG) frequency ranges, as well as electromyogram (EMG) signal morphology to identify CA; however, because of a lack of reliability in arousal detection, even among well-trained human scorers employing the AASM rules, CA has had limited efficacy in describing healthy sleep and/or in diagnosing sleep pathology. The purpose of this study is to increase the reliability of CA detection utilizing Power Spectrum Density (PSD). The exact frequency bands needed for CA detection for each sleep stage will be identified. It will be tested whether or not, the submental activity is necessary in slow wave sleep (SWS) to increase the reliability of CA detection. Methods: Previously recorded 30-second EEG sleep epochs from healthy adult subjects (N = 99) were examined in this study. The average and standard deviation of the relative powers of all frequency bands (Delta, Theta, Alpha, Sigma, Beta1, Beta2, and Gamma) were computed for all epochs and sorted by sleep stages. Using EEG activity, the relative power of specific frequency bands was compared to the average plus a multiple of the standard deviation for the purpose of detecting CA in each sleep epoch. EMG activity was also included for all sleep stages. The average and standard deviation of the relative amplitudes were then computed for all epochs and sorted by sleep stages. For each sleep stage, CA detection was achieved by comparing the standard deviation of the amplitudes to a multiple of the averages of the standard deviations. This experimental scoring technique was then compared with EEG epoch data scored by the Sleep Heart Center Study (SHHS) sleep scientists. An estimate of reliability was obtained using the Cohen kappa, sensitivity, and specificity measures. Results: Optimum CA detection entailed using a combination of different explicit frequency ranges for different respective stages. Based on the reliability calculated from the Cohen kappa, the optimum frequency bands for stage1 were: Beta1 (16-24 Hz), stage2: Beta1 (16-24 Hz), stage3: Beta1 and Gamma (24-48 Hz), stage4: Beta1 and Gamma (24-48 Hz), and stage REM: Delta, Alpha, Beta1, and Gamma (0-4, 8-12, 16-24 and 32-48 Hz). It was also found that the use of EMG for NREM sleep stages increased the sensibility. The corresponding statistical measures for all sleep stages were: Sleep Stage Cohen kappa Sensitivity Specificity Stage1 0.16±0.025 63±3% 56±2% Stage2 0.35±0.025 61±3% 81±1% Stage3 0.48±0.050 60±5% 96±1% Stage4 0.73±0.170 73±15% 98±2% Stage REM 0.48±0.030 58±3% 93±1% Conclusion: Careful consideration of the frequency band is necessary in order to increase the reliability of the detection of CA in EEG. Submental activity was also found to increase the reliability of CA detection in sleep stages SWS. Moreover, incorporating submental activity similarly increases most statistical measures. In conclusion, the AASM rules for detecting CA in EEG would be improved if the criteria included specific frequency ranges and submental activity in SWS sleep stages.