Researchers at Children’s Hospital of Philadelphia have developed a machine learning tool designed to better tell when a child is asleep and when they are simply lying still.
Children’s sleep outside the lab was commonly measured with actigraphy, which used a wrist-worn motion sensor.
However, this approach also had a well-known weakness: quiet wakefulness often mistaken for sleep. A child reading in the dark or staring at the ceiling barely moved and the device can log that time as rest.
What Did the Study Reveal?
The research team trained seven machine learning models on data from children revealed the study describing the work.
They then tested how well each model separated sleep from wakefulness in a lab setting, where overnight sleep studies provide a reliable benchmark.
The models were compared against existing scoring methods widely used in research, as well as a classifier trained on adult data. Sixty-five children took part in an overnight sleep study, each wearing a sensor on the non-dominant wrist.
Of the methods tested, a long short-term memory (LSTM) classifier trained on pediatric data, a type of model that is known to retain key information over long stretches of time, was best at distinguishing sleep from wake.
It reportedly outperformed both traditional sleep-scoring algorithms and a machine learning classifier trained on adult data. The results underscore the need to build and validate wearable-device algorithms specifically for children.
The research was reportedly funded by the Eunice Kennedy Shriver National Institute of Child Health and Human <a href="https://myappsplus.com/holiday-extras-leverages-codex-for-faster-app-development-and-growth/” title=”Holiday Extras Leverages Codex for Faster App Development and Growth”>Development, part of the National Institutes of Health, under Award Number R01HD100421.
What It Could Mean for Families
Better sleep tracking matters because sleep affects children’s attention, mood, learning and physical health.
Accurate measurement could help doctors and researchers understand sleep problems, including how long it took a child to fall asleep and how often they wake overnight.
Because the tool used data from wearable devices, it could eventually allow sleep to be monitored in a child’s own bedroom rather than only in a sleep lab.
What Were the Limitations of the Study
Experts caution that lab results don’t always carry over to real life. Children’s sleep at home can be messier, with naps, restless nights and devices that slip off or are not worn consistently.
Another problem was that many sleep-scoring algorithms were built using adult data. Children’s sleep differs from adults’ in timing, depth and movement, so tools trained on grown-ups can miss the mark when applied to kids.
The researchers said further testing will be needed before the tool can be used widely.
