健康与遗传心理学研究室知识库

目的:1.利用恢复时段HRV信号对任务初始时段急性应激的响应验证恢复时段HRV信号对任务初始时应激状态具有回溯能力;2.基于恢复时段HRV对早期任务状态的回溯能力建模识别任务初期应激状态。

方法:本研究以蒙特利尔应激诱导任务作为实验室应激诱导源，随机选取133名大学生参加实验，采集个体在三个实验状态下:基线时段(A时段，心算前的连续2min，任务时段(B时段，心算任务时段，4min和恢复时段(R时段，心算结束后的连续2min)的HRV特征，使用心算任务开始时被试自评的“开始是否很紧张”标签和心算任务结束之后被试的即刻自评的“结束后是否比开始感到轻松”标签分别作为A时段应激状态(S1状态)和R时段应激状态(S2状态)， S1和sS2状态均分为高应激状态和低应激状态。采用Lasso回归分别筛选出对S1状态和S2状态具有预测作用的R时段HRV特征，选择具有预测作用的不同时段HRV特征差值和差值状态D_status(包括D1, D2, D1表示R时段与A时段的差值状态，D2表示B时段与A时段的差值状态)作为自变量，性别作为控制变量，S1状态作为因变量构建混合效应模型，利用多重比较方差分析检验HRV差值特征与D_ status之间是否存在交互作用以此验证HRV特征对S1状态的回溯能力;最后利用弹性网络对R时段与A时段的差值特征作特征筛选并选用随机森林算法和支持向量机算法对S1状态做识别探索。

结果:1.R时段hr-var, SDANN, LF/HF对4分钟前的应激状态有预测作用，R时段的hr_ var和vec_ len对当前时段的应激状态具有预测作用;LF/HF和SDANN的特征差值与差值状态D_ status存在交互作用(p<0.05，从而证明这两个参数对4分钟前的应激状态的具有回溯能力;;2.利用弹性网回归筛出D1下的hr_ var, SDANN,LF/HF, HF_por,vec_ len,SD2和hist -width分别投入到随机森林和支持向量机模型对S1状态进行识别，采用100次五折交叉验证的方式获取平均准确率和ROC曲线，随机森林识别正确率达到75.4%, AUC面积为0.76 。

Objective: 1.The responses of the HRV signal during the recovery period to acute stress in the initial stage of the task are used to verify its backtracking abifity for the initial stress state of the task; 2. Modeling and recognizing the initial stress state based on the backtracking characteristic of the HRV during recovery period w.r.t. the early task state.

Method: In this study, the Montreal pressure induction task was used as the stress inducement in the laboratory. 133 college students were randomly selected to take part in the experiment. The data of HRV features under three experimental states were collected, including the baseline period (A period, the continuous 2 minutes before the mental calculation), the task period (B period, mental calculation task period, 4 minutes) and the recovery period (R period, the continuous 2 minutes after the heart calculation).Labels corresponding to the self-evaluations with the content of "Are you stressful now?”in the beginning of the mental calculation task and "Do you feel more relaxing compared with your feeling in the beginning?”are deemed as the responses of the A period stress state (S 1 state) and the R period state (S2 state). S1 and S2 state were divided into high stress state and low stress state. The lasso regression was employed to select the HRV features during the R period that had predictive effects to the state of S 1 and S2.Using the differences of predictive HRV features among difference period and the state of difference (D_status, a variable conclude D1 which expressed the difference of the R period and A period, D2 which expressed the difference between the B period and the A period) as the independent variables, the sex attribute as the control variable and the S 1 state as the dependent variable to construct a mixed effect model.The type II Wald chi square analysis was used to test whether there are interactions between the HRV difference features and the D_status, so as to verify the backtracking ability of the HRV features to the S 1 state. Finally, the differences of HRV features during the R period and the A period were sifted by an elastic network model. The random forest and support vector machine algorithm are used to identify the S 1 state.

Result: 1 .The hr var, SDANN, and LF/HF features have predictive effects to the stress state in early 4 minutes and the hr_ var and vec_ len features have predictive effects to the current stress state during R period; The difference between LF/HF and SDANN and D_statur have interactions (p<0.05), which proves that these two parameters have the backtracking ability to the stress state in early 4 minutes; 2.The hr_ var, SDANN, LF/HF, HF_por, vec_ len, SD2 and hist width sifted by the elastic network are used as the input features for the random forest and the support vector machine model to identify the S1 state respectively. The average accuracy rate was obtained by 100 times five-folds cross validation. The accuracy rate of random forest reached 7_5.4% and the corresponding AUC area is 0.76.