Research On Depression Recognition Based On Eye Movement Characteristics

As a common mental disorder,depression has no objective criteria for its accurate assessment and diagnosis.The eye is the person's first sense organ,across the psychology research which may reveal the person's inner mood.Based on the experimental data of eye movement in psychology,this paper objectively qua ntifies the abnormal eye movement of depressed people by processing the characteristics of eye movement and constructs the recognition model of depression by using machine learning method to assist doctors in clinical diagnosis.The main contributions of this thesis are summarized as follows:(1)Use different methods to construct two kinds of data sets based on the eye movement data: eye-tracking trajectory data set and eye movement feature data set.The first data set is in order to extract the eye-tracking trajectory features from video data and text data in the collected data by multi-modal fusion method to construct eye-tracking trajectory data set.The second data set is in order to construct eye movement characteristics data set by analyzing the original data of eye movement characteristics and different experimental tasks of eye movement.(2)Recognition model of depression is based on eye-tracking trajectory characteristics.The characteristics of eye-tracking trajectory data in eye-tracking trajectory data set were analyzed to explore the differences between depressed and non-depressed people in eye-tracking trajectory data.We establish a neural network model suitable for eye-tracking trajectory data set by studying the characteristics of the data set and comparing three different network models by optimizing the network structure to improve the recognition rate of the model.(3)Recognition model of depression is based on eye movement characteristics.In this paper,we used spatial statistics,medical statistics and time series to analyze eye movement feature data in eye movement feature data set and explore the differences between depressed and non-depressed people in eye movement feature data.Through studying the characteristics of the data set,we use the random forest algorithm for feature selection,use the XGBoost classification model to establish a depression recognition model based on eye movement characteristics,and evaluated the model according to different tasks of the eye movement experiment.Additionally,the XGBoost classification model selects the random forest classification model and the Bi LSTM classification model for comparative analysis from two aspects: the different evaluation indexes of the same eye movement experimental task and the different eye movement experimental tasks of the same evaluation indexes.The experimental results show that the recognition rate of the depression recognition model reached 83.17% in the experiments on eye-tracking trajectory characteristics.And the depression recognition model reached 84.92% in the experiments on eye movement characteristics.Meanwhile,in the comparison experiment between models and tasks,the XGBoost classification model shows the best results.