We are pleased to annonce the publication of our work “Deep Learning for Pediatric Sleep Staging from Photoplethysmography: A Transfer Learning Approach from Adults to Children” in IEEE Transactions on Biomedical Engineering.
Sleep staging is an essential component in the diagnosis of sleep disorders and management of sleep health. It is traditionally measured in a clinical setting and requires a labor-intensive labeling process. Recent data-driven algorithms for sleep staging, using the photoplethysmogram (PPG) time series, have shown good performance in adults (2020 Sridhar et al., 2022 Kotzen et al.) However, for children, such algorithmic development has been very limited and is hindered by the lack of pediatric data availability with only a single important open dataset called CHAT. The lack of additional datasets prevents from training deep learning models on a very large amount of data so to improves performance. It also prevents from evaluating generalization performance on an independent external test dataset (target domain).
In this research, we hypothesize that a representation learned in an adult domain can be leveraged, through transfer learning and fine-tuning, to a children domain. The resulting algorithm performance was compared to the same model but without pretraining on adult data. The algorithms are evaluated on CHAT-test as on newly created external dataset of 825 PSG recordings from the Ichilov Sieratzki-Sagol Institute for Sleep Medicine.
This research makes the following key contributions:
The development of a robust deep learning model for the task of sleep staging (Wake, REM, NREM) in children.
Evaluation of generalization performance of this algorithm on a new external dataset;
Thorough quantitative error analysis to pinpoint the main reasons for misclassification.
Quantitatively, the figure below shows the test sets performance of the new training approach ("SleepPPG-Net (Adult->Children)", yellow curve) versus benchmark training approaches (red and blue curves).
Our findings establish a new state-of-the-art performance for the task of sleep staging in children using raw PPG. The findings underscore the value of transfer learning from the adults to children domain.
Comentários