Research On Depression Recognition Based On Eye Movement And EEG Data

With the development of the society, psychological illnesses, especially depression is increasingly emerging, seriously affecting people’s physical and mental health, which has very bad effects on both individuals and society. Therefore, the early detection of depression, namely the recognition in the early stage of mild depression, is very important to the prevention and treatment of depression. At present, researches using eye tracking method and EEG(Electroencephalograph) are developing, and applied more and more in the fields of psychological researches and medical diseases researches. Eye movements can reflect the external performances of inner activities of human, and EEG signals are objective physical signals, which can reflect the cognitive activities of human relatively truly.People in a wide range of ages may be depressed, and the research methods should be different according to different ages, different backgrounds, and so on. Therefore, we only chose mild depressed college students as our subjects in this study. In this paper, we combined the eye tracking techniques and EEG recording techniques, and designed three experiments to study the differences between mild depression college students and normal students. By collecting eye movement data and EEG data synchronously, and combing with classification algorithms, we studied the recognition of mild depressed college students. By analyzing the results of classification, we tried to discover the classification algorithms which had relatively high and stable classification accuracy. Selected by Beck Depression inventory(BDI), 28 college students among all the freshmen and sophomores in School of information Science and Engineering were invited to be our subjects, of whom a certain proportion were mild depression students. After data collecting, we dropped the invalid data, and finally used data of 22 subjects to do data analysis.Firstly, we did classification on eye movement data collected during the passages reading tasks, facial expression pictures watching tasks, and smooth pursuing tasks separately. The classification results showed that, compared with the data collected during calm passages reading and calm pictures watching, eye movement data collected when subjects read passage with emotional bias and watched contrast pictures of calm pictures and emotional pictures had higher classification accuracy, and eye movement data collected during contrast pictures watching tasks had better classification results than that collected during emotional passages reading tasks. Classification accuracy of eye movement data collected during smooth pursuing tasks was relatively lower than that collected during passages reading tasks and pictures watching tasks. Besides, simple recognition tasks could not distinguish mild depression people and normal people.Secondly, we did classification on EEG data collected during the facial pictures watching tasks. Preprocessing of original EEG signals included the procedures of filtering, segmenting, and feature extracting. We chose to study Alpha rhythm waves and Beta rhythm waves which appeared when people were awake, and extracted EEG features, Approximate Entropy(ApEn) and the Largest Lyapunov Exponent(LLE), from the two rhythm waves separately. Then we did classification on ApEn, LLE, and the combination of ApEn and LLE. The classification results showed that DTNB algorithm had relatively stable classification results for data collected during the contrast pictures watching tasks.Lastly, We associated the eye movement features with EEG features into new data sets, and did classification analysis on these data sets. The classification results showed that the classification results of the combination of eye movment data and ApEn and LLE from contrast pictures watching tasks were very stable, and the classification results using DTNB algorithm were all 71.11%. This illustrated that data of ApEn, LLE combined with eye movement features from contrast pictures watching tasks had good effects on the recognition of mild depressed people.In all, the combination data sets of eye movement features and EEG features from contrast pictures watching tasks had better classification results. The results of our analysis also showed that, for the combination data sets of eye movement features and EEG features, DTNB algorithm had higher classification accuracy and its classification results tended to be stable.