AI in measurement of sleep in pediatic populations - Umaer Hanif

Artificial intelligence (AI)-based sleep staging is increasingly used in sleep research and is beginning to enter clinical workflows. However, most automated algorithms have been developed using adult data, and their performance in pediatric populations, where sleep physiology, electroencephalography (EEG) characteristics, and recording conditions differ, remains insufficiently characterized.

In this presentation, we review the performance and clinical relevance of six widely used open-source automated sleep staging algorithms evaluated in two pediatric cohorts: adolescents undergoing full polysomnography and children recorded with simplified home-sleep EEG setups. Results will be discussed with a focus on clinically meaningful outcomes, including reliability of sleep stage classification, agreement with standard sleep metrics (total sleep time, sleep onset latency, wake after sleep onset), and robustness to reduced electrode montages and signal quality variations.

Overall, several algorithms showed encouraging generalizability to pediatric recordings, with two models demonstrating the most consistent performance across datasets. Nevertheless, stage N1 sleep remained difficult to classify, and accuracy declined with limited electrode configurations, issues that are particularly relevant for home monitoring and clinical screening applications.

These findings suggest that AI-based sleep staging may soon support large-scale pediatric sleep assessment, research studies, and potentially clinical follow-up. However, broader validation in diverse pediatric clinical populations and careful consideration of clinical context remain essential before routine clinical adoption.

The talk will highlight practical implications for clinicians, current limitations of AI sleep scoring in children, and future directions for integrating automated sleep analysis into pediatric sleep medicine.