Caracterización de la relación entre la actividad eléctrica cerebral y cardíaca en pacientes pediátricos con síndrome de apnea/hipopnea durante el sueño empleando medidas no lineales de causalidad
Jaimes Albarracin, Anggie Daniela | 2020
Sleep disorders are conditions that cause changes in normal sleep patterns and that can have consequences on human health both in the short and long term. These are grouped depending on their causes and conditions, being sleep apnea / hypopnea syndrome (SAHS) part of respiratory disorders.
SAHS is a disorder characterized by the presence of several respiratory pauses due to blockage of the airways during sleep causing a decrease in circulating oxygen. The standard treatment method for pediatric SAHS is adenotonsillectomy, a surgery that, according to various studies, has shown high effectiveness. This eliminates oxygen desaturations and contributes to the return of normal functions of the autonomic nervous system and cardiac activity.
Currently, and due to the alterations in the electrical activity of the brain and the heart caused by SAHS, interest has grown in the use of polysomnography to characterize the interaction between the electrical activity of the brain and the heart. The quantification of these topological characteristics is commonly done using Granger Causality (GC). This measure is a statistical notion of causal influence, based on the prediction through vector autoregression. However, in the last decade, alternative proposals have been presented to obtain similar measurements, based on neural network architectures, to find non-linear relationships in said causal influence.
This document compiles the approach of the project in which it was sought to characterize the relationship between cardiac and cerebral electrical activity in pediatric patients with SAHS in pre- and post-treatment stages. The project is divided into four stages. The first corresponds to the preprocessing of the signals, where at the end of this stage eight time series of normalized powers were obtained from each polysomnography record, represented in five EEG subbands (α,β,δ,θ,γ) and three of the heart rate variability (VLF, LF, HF). In the second, the calculation of Granger causalities was performed using two methods: one linear using the MVGC toolbox and the other non-linear using the development of artificial neural networks. Subsequently, in stage three, McNemar's Chi-Square statistical test was performed, with which it was possible to identify each of the pairs of subbands in which the treatment induced significant changes. Finally, in stage four, the corresponding analysis of the results obtained and a subsequent comparison of both methods was carried out, taking into account their advantages and limitations.
Finally, the results showed that the treatment allowed the recovery of connections mainly in the heart-brain and brain-brain interactions, where the gamma and delta subbands presented a high level of restoration. Additionally, it was evidenced that adenotonsillectomy not only promotes a general topology similar to that of healthy subjects, but also affects some of the interactions analyzed, generating alterations that were not previously found, as it was in the case of the heart-heart network. Finally, a noticeable difference was evidenced between the methods used, where the non-linear analysis yielded complementary results and better identified interactions reported in previous studies compared to the linear methodology.