| 其他摘要 | Depression is an emotional disorder that affects the body and mind of patients. Itis usually measured by self-report method or diagnosed by psychologists. Thosemethods also have some shortcomings, such as long time consuming, not suitable forrepeated measurement and leading negative emotions. Therefore, these methods havelow ecological validity in some scenes of identifying depression feelings, depressivesymptoms and depression. Based on this, this study explores the ecological recognitionmethod of depression based on facial behavior analysis, so that it can be used as analternative method when the traditional depression recognition methods are notapplicable.
In view of the low ecological validity of the existing automatic depressionidentification methods, the inability to evaluate the discrimination of the model for themulti-dimensional model and the insufficient interpretability of the models, this studydesigned a suitable ecological perception method respectively for the three depressionmeasurement scenarios. Learnt from the reliability and validity test method of multi-dimensional scale, this paper put forward the reliability and validity test method ofmulti-dimensional model, and used statistical analysis method to explore the facialexpression pattern of depression. This paper designed and implemented three studies toexplore the feasibility and effectiveness of the ecological perception technology ofdepression based on facial behavior. The results show that:
1) In large-scale screening scene, it is feasible and effective to conduct ecologicalperception of depressive feelings based on facial behaviors from ordinary self-introduction videos. The results showed that the prediction effect of facial and gazefeatures was higher than that of only facial features. In all of the models we tried, theridge model with a periodic kernel showed the best performance. The model showed amutual fund R-squared (R2) value of 0.43 and a Pearson correlation coefficient (r) valueof 0.69 (p < 0.001). In all facial muscle activities, AU20, AU9, AU10, AU4 and AU14were most correlated with the degree of depressive feelings.
2) In the group monitoring scene, it is feasible and effective to conduct ecologicalperception of depression symptoms using facial key point activities when people readneutral text, which were extracted by Kinect. The study used six factors of SCL-90 toestablish six ecological perception models of psychological symptoms. The correlationcoefficients between the predicted values and actual scores were 0.26 to 0.42 (P < .01),which indicated good criterion validity. All models except depression had highconvergent validity but low discriminant validity. Results also indicated good levels ofsplit-half reliability [from 0.516 to 0.817] (P < .001).
3) In the clinical scene, it is feasible and effective to conduct ecological perceptionof major depression using facial movements, which were extracted by Kinect. Theresults of Multivariate ANOVA showed that the emotional polarity of reading materialshad little effect on facial movements, while gender and diagnostic results had greatereffect on facial movements. The accuracy, sensitivity and specificity of male depressionperception model were 0.8, 0.81 and 0.78 respectively; The accuracy, sensitivity andspecificity of female depression perception model were 0.81, 0.88 and 0.69 respectively.At the same time, the test results of the models on other data sets showed that the modelshave good robustness.
The above researches showed that the ecological depression recognition methodbased on facial behavior is suitable for different depression measurement scenes, and itis feasible and effective. This paper creatively designs different ecological perceptionschemes according to different application scenarios, improving the ecological validity;evaluates the model with the reliability and validity test method of the scale, providinga feasible method for multi-dimensional model evaluation; and explores the facialexpression pattern of depression with the statistical analysis method, providing basisand reference for facial feature extraction of depression recognition in the future. |
修改评论