Mutual Information and Delay Embeddings in Polysomnography Studies

In polysomnography, multiple biosignals are acquired to conduct human sleep studies. The sensors and measurement methods correspond to qualitatively different communication channels and therefore represent integrative perspectives of the sleep dynamics. Many pathologies associated with sleep disorders can be studied ranging from sleeps apneas, parasomnias and bruxism to other neurological disorders like epilepsy or Parkinson’s disease. The polysomnography study is complex and require highly specialized expert analysis. In this work we use data from sleep apnea patients. Given the nonlinear nature of polysomnography biosignals, correlation based methods are not best suited for signal study or classification. A non-linear time-series analysis is performed on the polysomnographic data, using time delayed mutual information and delay-embeddings. Important characteristics of the respiratory dynamics are estimated by this procedure that enables signal comparison and parameter selection for the design of predictive models and machine learning algorithms.