EEG Signal Modeling Method Fused With Demographics And Its Application In Depression Recognition

With the rapid development of economy and the improvement of material living standards,people's life rhythm and social competition are increasing.Physical,psychological and social pressures are also increasing.The incidence of mental illness has been rising continuously.And the increase of depression is especially obvious.Depression is a very complex and dynamically changing mental disorder,and its effective recognition often depends on the doctor's clinical experience and the patient's self-assessment scale.However,due to the subjectivity of doctors and the shortage of medical resources,an objective and effective way for depression recognition is urgently needed.Electroencephalograph(EEG)reflect the active state of brain nerve cells,which is difficult to disguise and easy to collect.In recent years,EEG-based depression recognition has become a hot topic in the development of biomedical engineering.However,the complexity and non-stationarity of EEG signals are two biggest obstacles to this application.In addition,the generalization of detection algorithms may be degraded owing to the influences brought by individual differences.Therefore,how to explore the more effective correlation and extract high-level representations that better reflect depression in EEG are important prerequisite for modeling a depression recognition model with stronger generalization.In view of the correlation between EEG signal and demographics,such as gender,age,etc.,and the correlation between these demographic factors and the incidence of depression,it would be better to incorporate demographics during EEG signal modeling and depression recognition.In view of this,a series of work are carried out around EEG signal modeling method fused with demographics and applied to the objective and quantitative recognition of depression,which achieved better recognition effect.The main work and contributions are summarized as follows.1)This paper proposes an EEG signals modeling method fused with demographic auxiliary tasks.Explore more effective features in EEG signals through convolutional neural networks(CNN).Incorporate gender and age information into neural networks through multi-task.Expand the learning goals of neural networks.This method uses the superior performance of representing learning to explore the correlation between depression and gender,age in EEG signals.The performance of depression recognition model is enhanced.2)This paper proposes an EEG signals modeling method with demographic attention mechanism.First,the CNN is also used to extract EEG features.Then the demographic attention mechanism is proposed,which incorporates gender and age information into the EEG signals modeling.In this study,demographic attention mechanism can adjust EEG feature maps though attention weights and explore the correlation between EEG signals and demographics.The performance of depression recognition model is enhanced.3)Experimental results on 170(81 depressed patients and 89 normal controls)subjects showed that the proposed EEG signal modeling method fused with demographics are superior to the unitary CNN without gender and age factors.Among them,the demographic attention mechanism method has the best performance,and the accuracy can reach 75.3%.It was proved that gender and age factors are helpful in modeling EEG depression recognition.However,the correlations between multi-task are uncertain,and it may cause negative transfers.The representations of multi-task learning are unclear,which cause that the performance of multi-task learning is inferior to demographic attention mechanism.In summary,the EEG signals and demographics modeling method proposed in this paper can explore the correlations between EEG signals,depression and individual,then generate an effective representation for depression recognition task.It means that it is necessary to jointly model EEG signals and demographic information,which will improve the robustness and generalization of classification model.This paper provides a new idea for modeling depression recognition based on EEG signals.