【摘要】 目的:建立气道廓清动作的声学分类器,为实施肺康复的咳嗽训练提供监测工具。方法:健康男性11例和女性15例,分别在平卧位、45°靠坐位和90°端坐位,根据随机视觉指令执行咳嗽和清嗓动作各10次,并同时录制声音,分析声音片段的时域、频域和信息域特征,由此构建声音的特征矢量用于机器学习。采用的模型包括:线性判别分析、分类回归决策树、随机森林和线性分类器型支持向量机。结果:模型间比较显示随机森林方法所建分类器具有更高的准确度(0.9162)和一致性(Kappa值为0.8323)。验证结果显示该模型无论在区分体位因素或不区分体位情况下,对咳嗽音有较高的准确度、一致性、敏感度和特异度。结论:咳嗽和清嗓动作具有声学差异,并且这种差异可以由随机森林方法构建机器学习模型加以分类,由此为肺康复治疗中采用声学手段辅助判断气道廓清动作类型提供了依据。更多还原

【Abstract】 Objective: To establish an acoustic classifier of airway clearance related behaviors and to provide a monitoring tool for cough training in pulmonary rehabilitation.Method: Eleven healthy males and fifteen females were seated in flat,45-degree and 90-degree positions,respectively. Cough and throat clearing behaviors were performed 10 times each according to the random visual instruction,and the sound was recorded at the same time. For each sound segment,acoustic features of timedomain,frequency-domain and information-domain were analyzed,then extracted the feature vectors of sound for machine learning. The models contain linear discriminant analysis, classification regression decision tree,random forest and linear classifier type support vector machine.Result: The comparison between the models showed that the classifiers established by random forest method had higher accuracy(0.9162) and the Kappa value(0.8323). Whether or not the influences of postural factors are included,the verification results showed that random forest model had the highest accuracy,the Kappa value,sensitivity and specificity for classification.Conclusion: There are acoustic differences between cough and throat clearing,and the differences can be classified by machine learning model. Our study provides a promising strategy for developing airway clearance behavior related classifier based on acoustic information.更多还原