Abstract: Objective　To study the application value of respiratory sound characteristic parameter analysis based on electronic stethoscope in pediatric pneumonia，and evaluate effectiveness of the zero-crossing rate and short-term energy parameters in identifying pulmonary crackles and wheezes. Methods　Utilizing an electronic stethoscope，we collected respiratory sounds from children hospitalized with pneumonia at The Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University from July 2021 to December 2022，extracted four feature parameters—maximum zero-crossing rate（zcrMax），minimum zero-crossing rate（zcrMin），maximum short-term energy（enMax），and minimum short-term energy（enMin）-to construct a model for the identification of pulmonary crackles and wheezes，and assessed the model performance via the receiver operating characteristic（ROC） curve，accuracy，sensitivity，and specificity. Results　In this study，a cohort of 5，157 respiratory sounds data entries were included. The maximum zero-crossing rate in Group A（crackles only） was significantly higher than that in the other three groups（P<0.001）. The maximum short-term energy in Group B（wheezes only） was greater than that in both Group A and Group D（normal breath sounds）（P<0.001）. A significant difference in the minimum shortterm energy was observed between Group B and Group C（both crackles and wheezes）（P=0.004）. Random forest model demonstrated higher accuracy compared to linear discriminant analysis，support vector machine，and Logistic regression model. The random forest model achieved an accuracy of 68.5% in identifying the presence of abnormal breath sounds，and its accuracy reached 75.2% for identifying the co-existence of both crackles and wheezes. Conclusion　It is feasible to use the zero-crossing rate and short-term energy as characteristic parameters to distinguish pulmonary crackles and wheezes. Constructing the random forest model using these characteristic parameters of breath sounds based on electronic stethoscope demonstrates certain advantages in the diagnosis of childhood pneumonia and provides a new technical approach for auxiliary diagnosis.

Key words:

摘要： 目的　探讨基于电子听诊器的呼吸音特征参数分析在儿童肺炎诊断中的应用价值，评估过零率和短时能量等特征参数对识别肺部干湿啰音的有效性。方法　利用电子听诊器收集2021年7月至2022年12月在温州医科大学附属第二医院育英儿童医院住院肺炎患儿的呼吸音，提取最大过零率（zcrMax）、最小过零率（zcrMin）、最大短时能量（enMax）和最小短时能量（enMin）4个特征参数，构建模型用于识别肺部干湿啰音，并通过受试者特征曲线、准确度、敏感度及特异度评价模型性能。结果　共纳入5157条呼吸音数据。甲组（仅湿啰音）的zcrMax高于其余3组（P<0.001），乙组（仅干啰音）的enMax大于甲组及丁组（正常呼吸音）（P<0.001），乙组和丙组（同时有干湿啰音）enMin差异有统计学意义（P=0.004）。随机森林模型在判断是否存在异常呼吸音的准确度为68.5%，同时存在湿啰音及干啰音的准确度达75.2%，优于线性判别分析、支持向量机和逻辑回归模型。结论　过零率和短时能量作为特征参数区分肺部干湿啰音具有可行性，构建基于电子听诊器呼吸音特征参数的随机森林模型在儿童肺炎诊断中表现出一定优势，为辅助诊断提供了新的技术手段。

关键词: