基于自然步态分析的抑郁和焦虑自动感知技术研究
文献类型：学位论文
| 作者 | 缪蓓蓓 |
| 答辩日期 | 2021-12 |
| 文献子类 | 继续教育硕士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 刘晓倩 |
| 关键词 | 抑郁 焦虑 心理健康 自动感知 步态分析 |
| 学位名称 | 理学硕士 |
| 学位专业 | 应用心理学 |
| 其他题名 | Research on Automatic Depression and Anxiety Perception Technology Based on Natural Gait Analysis |
| 中文摘要 | Mental health issues are already a global problem, with more than 350 million people worldwide suffering from depression, according to data released by the World Health Organization in 2021. It has previously been reported that depressive and anxiety disorders have become the number one and number six causes of non-fatal health loss globally, respectively, and the prevalence continues to increase each year. Low screening and consultation rates for depression and anxiety are prevalent in countries around the world due to the lack of health and medical resources, and traditional scale screening and one-on-one mental health diagnosis methods have shown limitations and inadequacies, so there is an urgent need to explore new methods and models for depression and anxiety perception and identification.This paper focuses on the automatic depression and anxiety perception technology based on natural gait analysis, which includes the following three aspects.(1) Gait behavior feature extraction technology for depression and anxiety recognition. The study used Patient Health Questionnaire (PHQ-9) and Generalized Anxiety Disorder (GAD-7) to assess the depression and anxiety scores of the subjects, and computer technology to detect and track the two-dimensional coordinates of body joints in the gait videos. Through correlation analysis, it was found that the motion fineness of some body joints was significantly and positively correlated with depression or anxiety, and the correlations were more pronounced in females; by the difference test in movement fineness of joint points on the upper and lower groups of depression or anxiety, it was found that there were significant differences in the motion of some joints of the right upper limb during walking between those with upper and lower scores of depression or anxiety. Based on this, gait temporal behavior features were extracted in the time domain and frequency domain, respectively, to describe the characteristics of gait motion variation of the individuals.(2) Study on the construction techniques of depression and anxiety score assessment models based on natural gait analysis. The cross-validation results showed that the correlation coefficient of the depression identification model was above 0.5, the classification accuracy was 86.4% and the AUC value was 0.754. The correlation coefficients of the anxiety identification model was above 0.4, the classification accuracy was 78.4% and the AUC value was 0.610. The results indicated that the models were valid and had practical values.(3) Study on the construction techniques of depression and anxiety classification models based on natural gait analysis. To meet the needs of identifying high-risk groups in depression and anxiety screening, the upper 27% and lower 27% of the scores were selected from the sample data set, and the machine learning classification algorithms were used to construct the upper and lower groups classification models for depression and anxiety, respectively. The classification precision of the depressed group was 0.71, the recall was 0.83, the F1 score was 0.77, and the AUC value was 0.67. The classification precision, recall and F1 score of the anxious group were 0.83, and the AUC value was 0.79. The results showed that the classification models could well distinguish the depression and anxiety high risk groups from the healthy ones.This paper explores the association between gait behavior and depression or anxiety through a data-driven approach, and initially constructs automatic identification models for depression and anxiety. Compared with traditional scale screening methods, this technique has the following advantages: (1) non-intrusive measurement methods with high ecological validity, no learning effect under repeated administration, and can be used for long-term follow-up monitoring; (2) low equipment cost, labor-saving, easy and convenient to use, time-saving and efficient, and suitable for large-scale primary screening scenarios. Due to the above advantages, the automatic depression and anxiety perception technology based on gait analysis can be used for daily monitoring and auxiliary diagnosis of mental health in primary care institutions, as well as for large-scale primary screening of mental disorders, and is expected to be an effective supplement to traditional diagnostic screening methods and be valuable in improving the screening and treatment rates of mental disorders. |
| 英文摘要 | 心理健康问题已经是一个全球性的问题，据世界卫生组织2021年发布的数据，全球有超过3.5亿人罹患抑郁症。此前曾有报告指出抑郁障碍和焦虑障碍分别已成为全球导致非致命性健康损失的首位和第六位原因，而两者的患病率仍在逐年增加。由于卫生和医疗资源的缺乏，抑郁和焦虑疾病的筛查和就诊率过低问题在全球普遍存在，而传统使用的自陈量表筛查方式和医生-患者一对一的心理健康诊断治疗方法已经显示出局限性和不足，亟待探索抑郁和焦虑感知识别新方法和新模式。本文重点围绕基于自然步态分析的抑郁和焦虑自动感知识别技术展开研究，具体内容包含以下三个方面：（1） 面向抑郁和焦虑识别的步态行为特征提取技术研究。研究使用患者健康状况问卷抑郁量表（PHQ-9）和7项广泛性焦虑障碍量表（GAD-7）评估被试抑郁和焦虑水平的得分，并利用计算机技术检测并跟踪个体步态视频中若干躯干关节点的二维坐标。通过相关分析，发现部分躯体关节点的运动精细程度与抑郁或焦虑水平呈显著正相关，且这种相关关系在女性身上表现得更为明显；通过关节点运动精细程度在抑郁或焦虑的高低组别上的差异性检验发现，抑郁和焦虑得分较高与较低的人群在行走过程中右侧上肢的部分关节点运动存在显著差异。在此基础上，分别在时间域和频率域空间提取步态时序行为特征，描述个体步态运动变化的特点。（2） 基于自然步态分析的抑郁和焦虑得分评估模型构建技术研究。利用机器学习回归算法分别构建基于步态特征的抑郁和焦虑状态得分评估模型，交叉验证结果显示，模型对抑郁识别的相关系数在0.5以上，准确率达到86.4%，AUC值达到0.754；对焦虑识别的相关系数在0.4以上，准确率达到78.4%，AUC值达到0.610；性能评估结果表明模型有效且具备实用价值。（3）基于自然步态分析的抑郁和焦虑高低组分类模型构建技术研究。为满足抑郁和焦虑筛查中对高风险人群的识别需要，从样本数据集中选取得分前27%和后27%的样本，利用机器学习分类算法分别构建抑郁和焦虑高低组分类模型，结果显示模型对抑郁高分组的识别精度达到0.71，召回率达到0.83，F1分数达到0.77，AUC值为0.67；在对焦虑高分组的识别精度、召回率、F1分数均达到0.83，AUC值为0.79，证明分类识别模型能较好地将抑郁和焦虑高风险人群与健康人群进行区分。 本文通过数据驱动的方式探究个体步态行为与抑郁和焦虑的关联关系，初步构建了抑郁和焦虑自动识别模型。相比传统的量表筛查方法，该技术具备以下优点：（1）无侵扰式的测量方法生态效度高，反复施测情况下不会造成学习效应，可用于长期跟踪监测；（2）设备成本低，节省人力，使用简单便捷，省时高效，适用于大规模初筛场景。基于上述优势，基于步态分析的抑郁和焦虑自动感知技术可用于基层医疗机构对心理健康的日常监测和辅助诊断，以及大范围的心理疾病初筛，有望作为传统诊断筛查方式的有效补充，在心理疾病的筛查率和治疗率的提高上发挥价值。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/45242]  |
| 专题 | 心理研究所_健康与遗传心理学研究室 |
| 推荐引用方式 GB/T 7714 | 缪蓓蓓. 基于自然步态分析的抑郁和焦虑自动感知技术研究[D]. 中国科学院心理研究所. 中国科学院大学. 2021. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。