EEG-based Analysis And Detection Of Brain Functional Network In Premenstrual Syndrome

Premenstrual syndrome(PMS)refers to a set of physical,emotional,and behavioral changes that occur in women of reproductive age on a cyclic basis prior to the onset of menstruation,and disappear with the arrival of menstruation.As a common gynecological disease with a high incidence rate,PMS has a insidious onset that is difficult to detect.By interfering with the normal functioning of the nervous system,PMS can cause problems such as emotional instability and cognitive impairment.In addition,PMS disrupts the excitability and inhibitory control of the cortical regions of the brain,resulting in defects in attention,working memory,and executive function,which seriously affect the normal life of patients.The neural mechanisms underlying the cognitive impairment caused by PMS are not yet clear.Electroencephalogram(EEG)signals reflect the changes in brain waves during activity and have the characteristics of high temporal resolution,making them ideal for revealing the characteristic changes in the brains of PMS patients during cognitive processes.Therefore,this study focused on PMS patients and healthy women as the control group,using the word-face Stroop paradigm in psychological experiments and32-channel EEG recording equipment to collect data.The collected data are analyzed using phase locking value analysis,and the brain functional networks of PMS patients and healthy women are constructed to explore the differences in the topological structure attributes of the two groups of subjects’ brain functional networks.The global and local parameters of the two groups of subjects’ brain networks are statistically analyzed using independent sample t-tests.This research provides new insights into the study of cognitive impairment in PMS patients and lays the foundation for the construction of corresponding graph neural network models based on network differences for the diagnosis of PMS patients.This thesis mainly carried out the following work:(1)The raw EEG data of PMS patients and normal controls are preprocessed and the changes in the mean phase locking value between electrodes are calculated for different frequency bands.A suitable threshold is used to construct the brain functional network,and the average clustering coefficient,shortest path length,global efficiency,and node degree attributes of the network are extracted for analysis to reflect the overall and local differences in the PMS patient’s network.During an emotional conflict task,abnormal connectivity is observed in the central region of the PMS patient’s brain network,where the network node degree is smaller than that of healthy women,resulting in a weakened ability to select emotional conflict responses and solve conflicts.Meanwhile,overall network properties are also abnormal: the clustering coefficient of the theta band in the PMS brain functional network is significantly reduced,reflecting a decrease in the internal connectivity of the brain and a reduction in the ability to process emotional information.The global efficiency and clustering coefficient of the alpha band brain functional network is significantly reduced,and the characteristic path length is significantly increased,indicating a decrease in the degree of grouping within the brain and a reduction in the synchronicity between brain neurons,resulting in a decrease in network information transmission capacity.In the beta band,there are significant differences between the PMS group and the HC group in terms of global efficiency and shortest path length,indicating that PMS can cause the brain to spend more time processing emotional interference signals.The mechanism of the decline in cognitive function in PMS patients is not only related to the decrease in control ability in the central region but also to the overall integration and information transmission capabilities of the entire brain network.(2)Through the analysis of brain functional networks,it is evident that the differences between brain networks in different frequency bands are more pronounced,and the distribution of EEG electrodes presents a non-Euclidean topology.This thesis proposes a prediction model based on a multi-frequency graph convolutional neural network.The input features are the topological structures of brain networks and differential entropy features from multiple frequency bands.Graph convolution is used to extract features,and attention mechanism is adopted to merge and fuse the features extracted from different frequency bands,considering that the contributions of the extracted graph features in different frequency bands are not the same.This model explores the spatial relationships of PMS patients’ EEG signals in multiple frequency bands and channels and achieves a high accuracy of 92.82% in PMS patient recognition,achieving good classification performance.(3)We have designed and implemented a premenstrual syndrome(PMS)brain network analysis and detection system.The system can automatically preprocess multiple commonly used raw EEG data formats.In terms of brain functional network analysis,the system provides various brain network construction methods from both linear and nonlinear relationships between nodes,facilitating the construction of different task-based brain functional networks for PMS patients and normal subjects and comparative analysis of network topology parameters.It also provides brain network and network parameter visualization functions.In terms of PMS detection,the system integrates a multi-frequency band graph convolutional neural network prediction model and provides other commonly used network models for PMS detection.This system provides great convenience for neuroscientists to study the differences in brain connections and has great potential for assisting in the diagnosis of PMS patients.