基于深度学习算法的抑郁症表情与动作及其心理整合变化的过程分析

Analysis of Facial Expressions and Movements Based on Deep Learning and Inner Emotional Experiences on a Patient with Depression

目的:探索基于深度学习算法的抑郁症患者表情动作变化与其症状转归及内在心理整合的关系。方法:采集1例抑郁症患者14次认知行为治疗的全程音视频资料,使用汉密尔顿抑郁量表(HAMD)与汉密尔顿焦虑量表(HAMA)评估病情,采用深度学习算法分析患者的表情熵、表情成分和动作幅度,采用整合分析方法评估患者的问题体验整合顺序(APES)。结果:治疗前后患者HAMD与HAMA减分率分别为90.91%和91.43%。随治疗进行,患者表情熵值呈现波动式上升,抑郁表情出现的概率总体呈降低趋势,整体动作幅度逐渐增高,语量逐渐增加。患者最初APES评分0分,治疗结束时APES评分为7分。结论:基于深度学习算法的抑郁症外显的表情与动作变化,与其内在的心理整合变化呈现一致性。将深度学习算法应用于抑郁症患者的表情识别和姿势评估是可行的。

Objective:Deep learning algorithms were used to explore the relationship between extrinsic bodily and facial expressions and intrinsic psychological assimilation of a depressed patient during the course of cognitive behavior therapy.Methods:The video data of the whole course of 14 treatments of a patient with depressive disorder were collected.Deep learning algorithms were used to analyze the patient’s expressions and movements,and meanwhile Assimilation Analysis were used to analysis the textual data transcribed from the audio.Results:The reduction rate of the patient’s HAMD and HAMA were 90.91%and 91.43%,in respective.The expression entropy increased from the beginning to the end of therapy.The expression components changed from mainly sad expressions to various expressions,as the treatment progresses.The movement range gradually increased.The results of analysis based on textual data showed that the volume of speech gradually increased,and the patient’s awareness and integration ability of their own problems gradually improved.Conclusion:The changes in the facial and bodily expressions of depression based on deep learning algorithms are consistent with the changes in their internal psychological assimilation.Deep learning algorithm to facial expression recognition and posture estimation of patients with depression is feasible.