Artificial intelligence accurately charts sleep stages without intrusive brain sensors

Researchers developed an AI model to track sleep stages using non-invasive sensors measuring heart rate, blood oxygen levels, and abdominal breathing instead of traditional brain wave sensors.
This method, published in the Journal of Sleep Research, aims to make clinical sleep tracking more comfortable and accessible for home use.
Traditionally, identifying sleep stages uses polysomnography, involving a night in a clinic with multiple wired sensors and brain wave tracking via EEG, which interrupts normal resting.
The AI model created by Serrano Alarcón and colleagues uses a U-Net deep learning algorithm and focuses on three physiological markers, showing 71% accuracy in sleep stage categorization on trained data and 66% on an external dataset.
The AI model’s secondary success lies in highly detailed, real-time tracking of sleep phases, second by second, allowing for more precise visualization of minor sleep disruptions.

You may have missed