An EEG Study Of Depression Based On The Modern Spectral Estimation

Depression, with obvious and long-term low mood as the main clinical features, is a common disease of emotional disorder harmful to human health. At present, the main method of diagnosing depression at home and abroad are clinical diagnosis and assessment scale. These methods can only take effect when reaching the diagnostic threshold, and are difficult to avoid interference factors such as subjective conceal of subjects, which easily lead to skew results. Burg et al. show that power spectrum in different rhythm of depression patients’ EEG has different characteristic with normal person. Therefore, we firstly screen depression patients and the normal by scale. Then, we collect the brain waves of those two populations, analyze EEG of depression patients and normal with modern spectral estimation method, and use Na?ve Bayes algorithm and K nearest node(KNN) algorithm to achieve accurate classification of two populations. Details are as follows.1. First, we screen people used for experiments by self-rating scale and MINI scale, and select 94 persons as subjects, with 47 persons forming depression group and others as normal control group. On this basis, using 3-lead generalize EEG acquisition instrument to collect brain electrical signals of two groups of subjects, as the original data of this study, in the resting state and under the audio stimulation.Then, using band-pass filter to filter the data, and further removing eye electricity,thereby clean EEG is obtained.2. For the clean EEG signals, we use AR model Burg algorithm in modern spectral estimation to estimate the power spectrum. On the basis, we extract absolute average power, gravity frequency, maximum power and power spectral entropy of EEG features, and use SPSS to make statistical analysis of the four EEG features. The results showed that there is a significant difference in terms of power spectrum between depression group and the normal group.3. Using SPSS to do independent sample T test for self-rating scale scores of two groups’ subjects. Test results showed that the scale scores between the two groups have significant differences. Health questionnaire scores(average 17.81) of thedepression group are significantly greater than the normal group(average 2.36).Doing Pearson correlation analysis for self-rating scale scores and the corresponding EEG features of two groups’ subjects. Analysis results showed that the health questionnaire and the maximum power have a good correlation(p = 0.02, r =0.338),which showed that we can verify the result of the determination from the perspective of the data.4. Selecting four EEG features to form an eigenvector, construct Naive Bayes classifier and KNN classifier, and then classify the two groups of subjects.Research results showed that the absolute average power and the maximum power of depression group were obviously higher than that of normal group. Because the Alpha wave power spectrum is inversely proportional to brain activity ability,brain activity of depression patients is weak relative to normal people. The health questionnaire and maximum power of EEG had a good correlation obtained by Pearson correlation analysis, which showed that we could quantitatively analyze the behavior characteristics of depression patients through the data. EEG features of two groups’ subjects form an eigenvector, and then input it into the classifier.Classification result was quite ideal, and the highest accuracy reached 70%. The results of this study provided a new path for the clinical diagnosis of depression,which can better explore the early warning theory and method of potential depression risk.