Development of a Multimodal Deep Learning Model for Pediatric Patients

Development of an Artificial Intelligence-Based Model for Assessing the Severity of Pediatric Obstructive Sleep Apnea

This study aims to develop a multimodal data-driven model integrating multiple noninvasive physiological signals to assess the severity of pediatric sleep-disordered breathing, using standard clinical sleep study results as the reference.

Pediatric obstructive sleep apnea may affect growth, development, cognitive function, and overall health. Although polysomnography is commonly used for clinical assessment, its application may be limited by time, cost, and accessibility. Recent advances in noninvasive monitoring technologies have provided new possibilities for sleep-related assessment. This study will collect and integrate multiple physiological signals from pediatric participants undergoing routine sleep examinations and to develop a data-driven model for evaluating sleep-related respiratory conditions. Clinical examination results will be used as the reference for model development and validation. The findings of this study are expected to support the development of a convenient and noninvasive approach for pediatric sleep assessment and may provide a reference for future clinical and home-based applications.

Inclusion Criteria:

• Individuals with clinical suspicion of obstructive sleep apnea who are referred for polysomnography

Exclusion Criteria:

• Intolerance to a fingertip or wrap-around pulse oximeter
• Presence of significant structural abnormalities of the upper airway
• Cardiac arrhythmia
• Neuromuscular disease
• Hospitalization within the previous one month