| dc.description.abstract | Sleep is a naturally occurring neurological state of the human body that helps restore and regenerate physiological and mental systems. Sleep comprises of AWAKE, Non-rapid Eye Movement (NREM), and Rapid Eye Movement (REM) stages. NREM sleep consists of four sleep stages N1, N2, N3, and N4. In an ideal sleep cycle, a human subject transitions through the sleep stages in the order AWAKE -> N1 -> N2 -> N3 -> N4 -> REM. Even though sleep is a resting state of the human body, physiological systems like the central nervous system, the cardiac system, and the respiratory system are still working in their vegetative state. However, the impact of sleep pathologies like sleep apnea on sleep stage transitions and connectivity between physiological systems during sleep remains largely unknown. This research presents a four-phased methodology to identify differences in sleep stage transition patterns and connectivity between physiological systems between control and sleep apnea subjects during sleep. The analysis is performed on polysomnography and histogram data collected from the Sleep Heart Health Study (SHHS) dataset. In phase I, the frequently occurring sleep stage transition patterns in healthy and unhealthy subjects are identified using the Apriori algorithm. In phase II, we studied the coupling strength and coupling direction between time series signals of brain wave activities measured as EEG waves in the δ, θ, α, σ, β, ɣ1, and ɣ2 bands. We proposed a framework that implements the Time Delay Stability (TDS) method that identifies the coupling strength between EEG bands and the LSTM-based Granger Causality (LSTMGC) estimation method that determines the coupling direction of the identified links. The results show a high coupling strength in control subjects in all sleep stages compared to sleep apnea subjects. Most links are bidirectional in the awake stage for control and sleep apnea subjects. However, in other sleep stages, more unidirectional links are identified in sleep apnea subjects, indicating a reduced coupling between EEG bands. In phase III, we developed an LSTM-based conditional Granger causality (LSTMCGC) method to identify the indirect influences of oxygen saturation ('sao2') and nasal airflow ('airflow') on brain-heart interactions during sleep. The results indicate that during light sleep, the sao2 and airflow signals have a low influence on brain-heart interactions in sleep apnea subjects but strongly influence the control subjects. In the REM sleep stage, the sao2 and airflow signals strongly influence brain-heart interactions for sleep apnea subjects and have a low influence for control subjects. In phase IV, we developed the Change in Causation during Sleep (CCS) model to study the changes in causation between physiological systems during an 8-hour long sleep. We mainly studied the causation between heart rate ('hr') and oxygen saturation signals. The overall results indicate a high causality from sao2 to hr signals in the REM sleep stage for sleep apnea subjects. But no such association is observed for healthy subjects. | |
