Abstract
Respiratory diseases, including pneumonia, asthma, bronchiolitis, and croup, remain the leading causes of pediatric morbidity and mortality worldwide. Diagnostic challenges persist, especially in low-resource settings lacking specialized tools. Artificial intelligence (AI)-based analysis of cough sounds has emerged as a promising, noninvasive diagnostic alternative. This systematic review synthesizes evidence on the predictive ability of AI algorithms for diagnosing specific pediatric respiratory diseases using cough sounds, evaluating their diagnostic performance, clinical applicability, and methodological quality. Following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) 2020 guidelines, six studies were included from 270 records identified in PubMed, Scopus, Web of Science, and IEEE Xplore databases. Eligible studies evaluated AI models such as logistic regression, convolutional neural networks (CNNs), support vector machines (SVMs), and hybrid feature-based approaches that combined acoustic and spectral features for disease classification. Techniques like wavelet-based feature extraction and late fusion, where outputs from multiple models are combined at the decision level, were reported to improve diagnostic accuracy. Sensitivity ranged from 82% to 94%, and specificity from 71% to 91% across studies, indicating high diagnostic potential, with some AI models outperforming conventional diagnostic methods such as the World Health Organization (WHO) clinical algorithms. Risk-of-bias assessment using Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) revealed concerns in four studies (67%), mainly due to retrospective designs, small sample sizes (ranging from 65 to 585 participants), and lack of external validation. Study limitations included heterogeneous outcome definitions and insufficient reporting of model calibration. Overall, AI-driven cough sound analysis demonstrates significant promise as a rapid, scalable diagnostic tool for pediatric respiratory diseases, particularly in resource-limited settings. Future research should focus on prospective multicenter validation, transparent reporting of methodological details and performance metrics, and integration into clinical workflows to ensure safe and effective real-world implementation.