Our systematic search of the Scopus, Medline, and Embase databases on 25 January 2023 identified six articles meeting the inclusion criteria. Quality assessment of the included studies was performed using the checklist for the assessment of medical artificial intelligence.
Our analysis shows variability in the input to the machine learning algorithms, such as the use of various cough sound features and combining cough sound features with clinical features. The use of the machine learning algorithms also varies from conventional algorithms, such as logistic regression and support vector machine, to deep learning techniques, such as convolutional neural networks. The classification accuracy for the detection of bronchiolitis, croup, pertussis, and pneumonia across five articles is in the range of 82-96%. However, a significant drop is observed in the detection accuracy for bronchiolitis and pneumonia in the remaining article.
The number of articles is limited but, in general, the predictive ability of cough sound classification algorithms in childhood acute respiratory diseases shows promise.
方法:我们对Scopus的系统搜索,Medline,和Embase数据库在2023年1月25日确定了六篇符合纳入标准的文章。使用医疗人工智能评估清单对纳入的研究进行质量评估。
结果:我们的分析显示了机器学习算法输入的可变性,例如使用各种咳嗽声音特征并将咳嗽声音特征与临床特征相结合。机器学习算法的使用也不同于传统算法,如逻辑回归和支持向量机,深度学习技术,如卷积神经网络。检测细支气管炎的分类准确性,臀部,百日咳,五篇文章中的肺炎占82-96%。然而,在其余文章中,细支气管炎和肺炎的检测准确性显着下降。
结论:文章数量有限,但总的来说,咳嗽声音分类算法在儿童急性呼吸系统疾病中的预测能力显示出希望。