This is a demo of an online service for automated analysis of nocturnal pulse oximetry in children aged 2-18 years. Machine learning (ML) models are used for Sleep and Wake staging, and to predict the presence or absence, and severity of sleep disordered breathing.

Learn more about the machine learning model used, including peer-reviewed validation studies.

Substantial emphasis is placed on model explainability to build clinician trust and confidence. Measures of uncertainty, global and local explanation methods, and data visualisation accompany ML-generated predictions. Furthermore, models use clinically relevant oximetric features that are familiar to sleep paediatricians.

Accepted uploads are exported oximetry data in a delimited text format. Most popular oximetry extraction and analysis tools allow for the exporting of oximetry data. Note: EDF files must be converted to text format.

No oximetry data on-hand? Sample oximetry data are available for you to try.

This demo is for academic and non-commercial purposes only. It is not approved for clinical use.

Models were trained and evaluated on pulse oximetry recordings of children aged 2-18 years. Performance cannot be guaranteed in children aged less than 2 years, or in adults.

Oximetry recordings sampled at 0.5Hz, with a Masimo SET averaging time of 2-4 seconds, were used for model training. Oximetry data sampled at a rate above or below 0.5Hz must be decimated or interpolated (with care), respectively. These processes are not performed through this service and must must be completed prior to upload. Oximetry recordings collected using prolonged averaging times may result in unreliable predictions.

Recordings using hospital-grade pulse oximeters from Masimo, Nonin, and Nellcor were used for model evaluation. Performance cannot be guaranteed on other oximeter brands. Analysis of data recorded using consumer-grade pulse oximeters is not recommended.