Related Experiment Video
Updated: Jul 9, 2026
06:37
Artificial Intelligence-Based System for Detecting Attention Levels in Students
Published on: December 15, 2023
PANDA pediatric arousal neural detection architecture.
Arnav Gupta1,2, Ayush Tripathi1,3, Wolfgang Ganglberger1,3
1Department of Neurology, Beth Israel Deaconess Medical Center, Boston, MA, USA.
NPJ Digital Medicine
|July 7, 2026
Summary
PANDA, a deep learning model, accurately detects pediatric arousals in polysomnography (PSG) recordings. This AI tool improves scoring consistency and reduces manual effort in sleep studies.
Area of Science:
- Artificial Intelligence
- Medical Technology
- Sleep Medicine
Background:
- Pediatric sleep studies, specifically polysomnography (PSG), are crucial for diagnosing sleep disorders.
- Accurate detection of arousals during PSG is essential for diagnosis but can be labor-intensive and prone to inter-scorer variability.
- Existing automated methods for arousal detection show limitations in consistency and accuracy, particularly in pediatric populations.
Purpose of the Study:
- To develop and validate PANDA, a novel deep-learning model for automated pediatric arousal detection in polysomnography.
- To assess PANDA's performance against established methods and human scorers, considering label noise and inter-rater reliability.
- To evaluate PANDA's generalizability across different cohorts and its potential to reduce manual scoring time.
Main Methods:
- Development of PANDA using a U-Net-style encoder-decoder architecture trained on 10-channel PSG signals.
- Training and validation on a large dataset of 9150 PSGs (7604 subjects) and 2455 PSGs (2000 subjects), respectively.
- Rigorous performance assessment using a platinum set of adjudicated PSGs, Cohen's kappa (κ), arousal index agreement, and external validation on CHAT and PATS datasets.
Main Results:
- PANDA achieved a mean subject-wise Cohen's κ of 0.45 on routine labels, significantly outperforming CAISR (κ=0.26).
- Agreement improved to κ=0.87 on a platinum set, indicating reduced scoring variability after expert adjudication.
- PANDA demonstrated superior event-wise agreement and lower bias compared to CAISR, with strong external validation results (κ=0.46 on CHAT, κ=0.38 on PATS).
Conclusions:
- PANDA represents a significant advancement in automated pediatric arousal detection from polysomnography.
- The model offers improved consistency and accuracy, addressing limitations of current methods and reducing the need for extensive manual scoring.
- PANDA has the potential to enhance the efficiency and reliability of diagnosing sleep disorders in children.
