A comparative study of four novel sleep apnoea episode prediction systems

The prediction of sleep apnoea and hypopnoea episodes could allow treatment to be applied before the event becomes detrimental to the patients sleep, and for a more specific form of treatment. It is proposed that features extracted from breaths preceding an apnoea and hypopnoea could be used in neural networks for the prediction of these events. Four different predictive systems were created, processing the nasal airflow signal using epoching, the inspiratory peak and expiratory trough values, principal component analysis (PCA) and empirical mode decomposition (EMD). The neural networks were validated with naïve data from six overnight polysomnographic records, resulting in 83.50% sensitivity and 90.50% specificity. Reliable prediction of apnoea and hypopnoea is possible using the epoched flow and EMD of breaths preceding the event.