Pediatric smart sleep diagnostics: integrating EEG, ECG, and respiratory signals for automated staging and OSA detection

Purpose This study aims to develop an automated system for pediatric sleep stage classification and obstructive sleep apnea (OSA) detection using machine learning techniques. The main research question addresses how EEG, ECG, and respiratory signals can be effectively utilized to enhance the accuracy and efficiency of pediatric OSA diagnosis, a condition that significantly affects children's health and development. Methods The study compares the performance of five machine learning algorithms on National Sleep Research Resource (NSRR) published dataset of polysomnographic sleep data collected from 425 children aged 6 to 12 years. EEG, ECG and respiratory signals undergo preprocessing and are divided into 10-second segments. Features are extracted using wavelet transform techniques. To optimize classification, sleep stages are re-categorized: N1 and N2 were grouped as Light Sleep (N1/N2) while N3 was retained as Deep Sleep (N3) for a four-stage classification. Results Ensemble models, particularly Random Forest integrated with XGBoost and AdaBoost, consistently outperformed other algorithms. In sleep stage classification, class balancing improved recall and F1-scores for minority stages (REM, N1/N2) while normalization modestly enhanced simpler models such as kNN and Logistic Regression performance. For OSA detection, Random Forest with AdaBoost yielded the best accuracy (89.49 %), demonstrating robust performance across apnea and hypopnea events. Conclusion The proposed multimodal Machine Learning framework effectively combines EEG, ECG, and respiratory features for accurate pediatric sleep staging and OSA detection. The findings highlight the clinical promise of ensemble learning in reducing manual scoring effort and enabling early diagnosis in pediatric sleep medicine.