In this study, we developed a system that identifies the respiratory effort related arousals (RERA) and the non-RERA, non-apnoea arousals by processing four signals of the chin EMG signal, two channels of EEG signals (C4-M1 and C3-M2) and the oximetry signal. The 2 EEG signals were processed identically. Firstly, preprocessing was applied to remove the baseline wander, unwanted low frequency components and the abrupt changes. Then, the EEG signals were divided into non-overlapping epochs and a power spectral decomposition was calculated resulting in 5 PSD features per epoch. The Chin EMG signals were processed in the same fashion and resulted in 5 PSD features. The artefact signal of the SaO2 signal was removed and the square root of the standard deviation of the signal was calculated. The features were combined into a 16-element epoch. Following this, features from surrounding epochs were combined with the current epoch. We compared the performance of combining features from one to four epochs either side of an epoch. The 10-fold cross validation results of three classifiers including linear discriminant analysis (LDA), logistic regression (LR) and single hidden layer feedforward neural networks (SHLN). The performance of our best system was an AUC 0.82 and an AUPRC of 0.24 using the 10 hidden units feed-forward neural network.
CITATION STYLE
Sadr, N., & De Chazal, P. (2018). Automatic Scoring of Non-Apnoea Arousals Using the Polysomnogram. In Computing in Cardiology (Vol. 2018-September). IEEE Computer Society. https://doi.org/10.22489/CinC.2018.252
Mendeley helps you to discover research relevant for your work.