Exploring precision medicine strategies for paediatric OSA: Construction of a multimodal intelligent system for surgical indication assessment

Abstract Purpose The prevalence of obstructive sleep apnoea (OSA) is high in children, for whom accurate diagnosis and timely intervention are crucial. Current single‐modal clinical evaluations have limitations. This study aimed to build a multimodal intelligent system to assist diagnosis and surgical decision‐making in children presenting with clinical symptoms. Methods A single‐centre retrospective study at Shanghai Children's Hospital enrolled 220 paediatric OSA patients between March and June 2025. Comprehensive multimodal data, including medical histories, nasopharyngoscopy images, and snoring recordings, were collected. Among them, 192 children were included in the surgical indication‑positive group and 28 in the surgical indication‑negative group according to the guidelines. To predict surgical need, a multimodal fusion model was trained on a split of 120 training and 100 test samples. Class imbalance was addressed using Focal Loss, inverse weighting, and minority augmentation with regularization. Results The multimodal fusion model demonstrated excellent performance, achieving a sensitivity of over 97% in identifying the necessity of adenoidal or tonsillar surgery. Compared to single‐modal and bi‐modal models, the multi‐modal fusion model showed substantial advantages in reducing both misdiagnosis and missed diagnosis, exhibiting strong screening capabilities. Conclusions The proposed system, integrating multimodal data, provides a sensitive and accurate tool to assist diagnosis and surgical decision‐making in this symptomatic population, offering a promising approach for improving clinical outcomes. Trial registration ChiCTR, ChiCTR2500098075.