Multimodal Assisted Diagnosis for Pediatric Respiratory Diseases Using Questionnaires, Cough Sounds, and Breath Sounds

This is a prospective, observational diagnostic accuracy study to be conducted at Shanghai Children's Medical Center. The study population will include children presenting with cough, wheezing, fever with respiratory symptoms, nasal congestion, rhinorrhea, sore throat, or other respiratory complaints, as well as healthy children recruited during routine health examinations. The study will establish a standardized and synchronized data collection workflow for pediatric symptom questionnaires, cough sounds, and breath sounds. All enrolled participants will complete a structured symptom questionnaire, undergo cough sound recording using a smartphone application, and undergo breath sound recording using an electronic stethoscope under unified protocols. Demographic and clinical information, including age, sex, disease duration, major symptoms, medical history, allergy history, family history, medication use, final clinical diagnosis, or health status assessment, will also be collected to construct a multimodal database of pediatric respiratory diseases including both disease cases and healthy controls. Based on this database, the study will develop a stepwise multimodal assisted diagnostic framework using a combination of conventional statistical learning and deep learning methods. Three diagnostic models will be constructed and compared: a symptom questionnaire-only model, a symptom questionnaire plus cough sound model, and a multimodal model integrating symptom questionnaires, cough sounds, and breath sounds. Using the final research labels determined by clinicians' diagnoses, health status assessments, and research team review as the reference standard, the study will evaluate the diagnostic performance of these models in distinguishing healthy children from children with respiratory diseases, screening for asthma and asthma-related cough, and identifying pneumonia, tracheitis/bronchitis, upper airway-related diseases, and common causes of chronic cough. Model performance will be assessed using AUC, AUPRC, sensitivity, specificity, positive predictive value, negative predictive value, F1 score, and accuracy. The study will further investigate the incremental value of cough sounds and breath sounds beyond symptom questionnaire information, and assess model stability and generalizability across different age groups, clinical settings, and device conditions. The findings are expected to provide evidence for the optimization, clinical translation, and potential home-based extension of multimodal artificial intelligence-assisted diagnostic models for pediatric respiratory diseases. The study will not interfere with routine clinical care, and the model outputs will not be used for real-time clinical decision-making.