Intelligent Voice Model: A New Paradigm Exploration for Child Health Management

The study employed a multicenter cross-sectional design. The real-world data collected for this study included normal and definitively diagnosed heart sounds in children with congenital heart disease, normal and definitively diagnosed respiratory tract infections in children with breath sounds, specific cough sounds, and normal and definitively diagnosed children's bowel sounds with diarrhea. The specialist team will carry out data governance, annotation, and feature sound extraction on the gathered normal and aberrant sounds, in order to generate a superior multimodal training dataset. Large model artificial intelligence algorithms (deep learning, machine learning, etc.) are used to model and train the algorithm model of the body voice AI recognition device, so that it can distinguish between normal and abnormal sound signals by AI. The results of body sound AI identification will be compared with diagnostic reports from echocardiograms, chest X-rays, and belly X-rays in terms of AUC (Area Under Curve) score, sensitivity, specificity, and accuracy to evaluate the impact of AI recognition devices on illness screening and supplementary diagnosis. External validation will be conducted using homogeneous data from other sites. This project aims to develop a new generation of intelligent sound auscultation instruments that could be used for early screening and auxiliary diagnosis of congenital heart disease , respiratory infections, diarrhea and other common multi-occurring diseases by utilizing large model artificial intelligence technologies.