应用啼哭信号分析方法评估婴儿术后疼痛的研究

发布时间：2018-05-26 02:11

本文选题：语音信号分析 + 啼哭　；参考：《上海交通大学》2014年博士论文

【摘要】：研究背景和目的寻找疼痛评估的客观指标一直是学术界难以解决的关键问题。尤其对于婴幼儿,因其发育尚不完善,无法进行疼痛的自身描述,导致在临床上缺乏简单、有效的婴幼儿疼痛评估标准。啼哭是多种婴幼儿疼痛行为学评估量表中均包含的特征之一,且有研究表明,婴幼儿疼痛与非疼痛的啼哭声存在不同。但是疼痛所导致的啼哭信号具体包含怎样的声学特征,且这种声学特征是否能作为评估婴幼儿术后疼痛的标准尚未见报道。我们拟通过提取6个月以内婴儿术后疼痛哭声信号特征,构建疼痛哭声识别模型应用于婴儿术后疼痛评估,以期为临床探索更客观、便捷的婴幼儿疼痛评估工具提供研究基础。研究内容和方法以6个月以内择期手术的婴儿为研究对象。记录婴儿手术前啼哭数据,以及从婴儿手术后进入麻醉苏醒室至婴儿苏醒后离开苏醒室这一时间段的啼哭信号。同时采用FLACC(Face,Legs,Activity,Cry,Consolability)评分量表对婴儿苏醒后的疼痛程度进行评估。以Adobe audition3.0对啼哭声音文件进行初步处理,通过Praat语音分析软件进行特征性参数分析;以MFCC(Mel Frequency Cepstrum Coefficient)参数作为特征性参数,以HMM(Hidden Markorv Model)模型作为训练模型,构建疼痛识别模型,同时以FLACC为标准,使用ROC(Receiver operating characteristic curve)曲线比较不同组合的HMM模型对婴儿术后疼痛啼哭的识别效能,寻找最优组合的疼痛识别模型,并检验该模型对术后疼痛以及重度疼痛的识别效能。使用SPSS 19.0进行数据处理和统计分析,计量资料以均数±标准差表示,数据均进行正态性检验和方差齐性检验,方差齐性的数据两组间比较使用t检验,方差非齐性两组间比较使用wilcoxon检验,多组间比较使用Kruskal-Wallis检验,p0.05差异具有统计学意义。使用R v 2.15.1统计作图。研究结果本研究为前瞻性研究(临床注册码:Chi CTR-OCH-14004648),共纳入155名择期手术婴儿。1.婴儿术后疼痛啼哭基频为654.7±195.8Hz,显著高于手术前啼哭基频464.6±146.1Hz,p0.01。2.术后疼痛组啼哭的基频(F0)、第一共振峰(F1),均方根(Root mean square,RMS)明显高于无痛/轻度疼痛组(p0.01),而音节间歇时间(interval between Syllables,IS)显著下降(p0.01)。3.组合为18state+12mixture数目的HMM模型对术后疼痛识别效能最大,其ROC曲线中的AUC(area under curve)为0.81±0.049,(95%CI:0.713-0.906),最佳截断点为0.558,敏感度为80.0%,特异度为77.1%。该模型诊断重度疼痛的AUC为0.764±0.059,(95%CI:0.648-0.880),特异度仅50.9%,但其敏感度非常高,为91.3%。结论婴儿术后的疼痛啼哭较非疼痛啼哭信号具有特征性的改变,这种改变可以被语音识别模型所探知,利用疼痛啼哭识别模型对婴儿术后疼痛的鉴别具有一定的诊断价值,本研究为进一步开发无创伤性、自动化的、有效评估婴幼儿术后疼痛的工具奠定了前期基础。
[Abstract]:Research background and purpose of finding the objective index of pain assessment is a key problem that is difficult to solve in the academic community. Especially for infants, the lack of self description of pain because of its imperfect development leads to the lack of simple and effective evaluation of infant pain in clinical. Crying is the assessment of many kinds of infant pain behavior. One of the features contained in the table shows that infant pain is different from that of non painful crying. But what is the acoustic characteristic of the crying signal caused by pain, and whether this acoustic feature can be used as a standard for assessing postoperative pain in infants and young children is not yet reported. We are going to extract babies within 6 months. In order to provide the basis for the clinical exploration of more objective and convenient infant pain assessment tools, a new model of pain crying recognition is used to provide research basis for the clinical exploration of a more objective and convenient tool for evaluation of infant pain. The crying signals of the waking room of the baby after the baby's operation were entered into the awakening room, and the FLACC (Face, Legs, Activity, Cry, Consolability) rating scale was used to evaluate the degree of pain after the awakening of the baby. The initial treatment of the crying sound file of Adobe audition3.0 was carried out by the Praat speech analysis. The software carries out the characteristic parameter analysis; taking the parameters of MFCC (Mel Frequency Cepstrum Coefficient) as the characteristic parameter, and using the HMM (Hidden Markorv Model) model as the training model, the pain recognition model is built, and the ROC (Receiver) Cepstrum curve is used to compare the infant model to the baby. The ability to identify the pain and cry after the operation, find the best combination of the pain recognition model, and test the recognition effectiveness of the model for postoperative pain and severe pain. Data processing and statistical analysis are performed using SPSS 19. The measurement data are expressed with mean standard deviation, and the data are tested in normality, variance homogeneity test and homogeneity of variance. The data between the two groups were compared with the use of t test, the two groups of variance inhomogeneous using the Wilcoxon test, the multiple groups using Kruskal-Wallis test, the P0.05 difference was statistically significant. The R V 2.15.1 statistics were used. The results of this study were prospective study (clinical registration code: Chi CTR-OCH-14004648), including 155 selected operation infants. The basic frequency of pain and crying after operation for infant.1. was 654.7 + 195.8Hz, which was significantly higher than the basic frequency (F0) and the first resonance peak (F1) in the pain group after p0.01.2., and the mean square root (Root mean square, RMS) was significantly higher than that of the painless / mild pain group (P0.01). The HMM model with a P0.01.3. combination of 18state+12mixture was the most effective for postoperative pain recognition, and the AUC (area under curve) in the ROC curve was 0.81 + 0.049, (95%CI:0.713-0.906), the best truncation point was 0.558, the sensitivity was 80%, the specificity was 0.764 + 0.059 for the 77.1%. mode diagnosis of severe pain. The degree of sensitivity is only 50.9%, but its sensitivity is very high. It is a characteristic change in the pain crying of 91.3%. conclusion after operation. This change can be detected by the speech recognition model. The use of pain crying recognition model has a certain diagnostic value for the identification of postoperative pain in infants. This study is a further development of this study. Traumatic, automated, effective tools for evaluating postoperative pain in infants have laid a foundation for early surgery.
【学位授予单位】：上海交通大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R726.1

【参考文献】