Prediction Of Dynamic Interaction Mapping Of Brain Waves Based On Deep Network Physiology

Brain waves are one of the most important physiological signals in the human body,and their dynamic interaction behavior is highly correlated with the generation of certain physiological states and functions.Therefore,the study of dynamic interaction mapping of brain waves has important physiological and medical values.A large number of scholars have extensively investigated the mechanisms of brainwave evolution and its physiological functions.However,there are still some unresolved challenges in this field,such as: 1.The dynamic interaction map of brain waves is highly correlated with a specific physiological state,and the study of this physiological map can provide a new perspective to understand the corresponding physiological state,but the existing deep learning methods cannot accurately model the complex endogenous mechanism of this map;2.The dynamic interaction map of brain waves is a unified body containing time domain,frequency domain and The existing interaction studies of physiological systems lack the exploration of implicit temporal relationships in the evolution of the atlas,and are unable to make accurate dynamic predictions of the atlas at future moments.In order to solve the above problems,this thesis addresses the problems of constructing dynamic interaction maps of brain waves,the internal temporal links of maps and the dynamic prediction of feature signals,respectively,with the following main contents:(1)To decode the complex endogenous mechanism in the interaction process for the scenario of brainwave interaction,a dynamic interaction quantification module is designed to quantify the interaction between brainwaves and construct an undirected weighted brainwave dynamic interaction map based on the synergistic burst mechanism between brainwaves and the time-delay stability theory in network physiology,and network physiology is combined with network physiology in the study of brainwave dynamic interaction map prediction.Deep network physiology is proposed for identifying,quantifying and predicting the endogenous mechanisms of the dynamic interaction map of brain waves by combining deep learning.(2)To address the problem that network physiology lacks the study of implicit temporal relationships in system interactions,a temporal link prediction framework is designed to simultaneously capture the time-domain,frequency-domain,and spacedomain features in the dynamic interaction map of brain waves and model the temporal relationships in the map structure.The GCN module is used to extract the single graph structure information in the dynamic mapping,and then the Bi LSTM module is used to model the implicit temporal relationships of the extracted structure information,and finally the WGAN-GP framework is used for adversarial training to generate higher quality topological snapshots to achieve autonomous learning and prediction of the EEG dynamic interaction mapping structure.(3)Since the the dynamic interaction map of brain waves is a unified body of feature signals and interaction relations of brain waves,the feature sequences on the nodes should be predicted in addition to the topological structure constructed through the interaction relations.Based on this,an informer-based time series prediction model of relative spectral power of brain waves is developed.First,the EEG recordings were intercepted by time windows and frequency decomposition and the spectral power of brain waves was calculated for each frequency band.Then,the relative spectral power is calculated to model the influence of each brainwave activity in the whole brain activity,and the relative spectral power time series is smoothed to capture the key information related to the internal evolution mechanism of the sequence,such as the quasi-steady-state behavior of brainwaves and the trend before and after the change of physiological state,finally,the smoothed brainwave normalized relative spectral power time series is modeled and predicted by the implicit time series relationship.The proposed methods have all been extensively experimented on real physiological datasets,and the results show that they can accurately predict the dynamic interaction map temporal link weights of brain waves with the dynamic interaction map node feature sequences of brain waves.The thesis has 40 figures,6 tables,and 81 references.