Study On Depression Recognition Based On Information Entropies Of EEG

Depression,as a commom psychological disease,seriously endangers people’s physical and mental health.With the acceleration of the modern social life rhythm,people often face tremendous mental pressure,which makes the incidence of depression is increasing trend year by year.Therefore,the recognition and diagnosis of depression is becoming more and more urgent.At present,the diagnosis of the clinical depression mainly relies on the qualitative methods,and the quantitative accurate diagnosis and treatment is rare.Establishing a physiological and psychological model of depression based on physiological data,exploring objective and effiective biomarkers,and studying the corresponding relationship between and physiological characteristics,could provide theoretical basis for realizing the quantitative and accurate diagnosis of depression.Based on effective biomarkers,combined with machine learning algorithms to build a depression recognition model to achieve quantitative,objective and precise diagnosis and treatment of depression,it has very important theoretical significance and clinical value for the accurate diagnosis of depression and improving the accuracy of depression recognition.Electroencephalogram(EEG)contains rich physiological and pathlogical information,and can reflect the overall electrophysiological activities of the brain and different brain functional states,making EEG shows unique advantages in the recognition of depression states.Studies show that depression in patients with frontal lobe activity has a significant change,at the same time in order to solve the process is relatively complex of EEG signal acquisition in previous studies.Thus,this paper collected the three-channel frontal EEG signal.Entropy feature matrices were established based on different EEG frequency bands.The statistical analysis and machine learning techniques are applied for evaluating the mapping relationship between the feature matrix and the different depressed states.The depression recognition models between different depression states and healthy control groups were designed and constructed by different classification algorithms,which can realize the effective recognition of different depression states.The main research contents of this paper are as follows:(1)EEG signals of depressed patients with different depressive states and healthy controls were collected.The signal is resampled and preprocessed by wavelet threshold transform and power frequency notch filter.A pure EEG signal for at least 5.5-minute was obtained.The Butterworth band-pass filter is used to extract four EEG frequency bands,and the entropy feature matrices of four frequency bands were extracted.The feature matrices based on different frequency bands were constructed to describe the changing laws of EEG signal,and to quantitatively characterize the complexity of EEG signals in patients with different depression states.(2)The statistical analysis was used to explore the distribution law between the entropy feature matrices and different depression states.According to the distribution results,the appropriate statistical testing methods were selected to analyze the difference of the entropy feature matrices under different depression states and the correlation between the feature matrices and different depression states.To explore the distribution law of entropy characteristics under different depression states.(3)The Relief algorithm is used to evaluate different feature matrices.The weights are assigned to the different features according to their importance,and the weighted optimal feature matrix is constructed.Depression recognition models based on random forest(RF),k-nearest neighbor(KNN),and support vector machine(SVM)are established,and the optimum parameters of the classifiers are estimated by 10-fold cross validation and grid optimization algorithm.One-to-one recognition of different depression states and healthy control group was carried out,and the average classification accuracy of the three depression recognition models was obtained.The results indicate that the depression state recognition model established in this paper can effectively identify different depression states.(4)The effects of EEG frequency bands in different depression states were analyzed.The results showed that the highest classification results were generated in the Alpha and Beta frequencies between the three groups of depressive states and the healthy control groups.The results indicated that the EEG frequency bands play an important role in depression recognition.The depression recognition model based on the optimal feature matrix and the optimal parameter in this paper can realize the effective recognition and classification between different depression states and healthy control...