Researchers from Mount Sinai’s Icahn School of Medicine, US, have introduced an AI sleep analysis tool, designed to process a complete night’s sleep data.
According to the US-based hospital network, the new patch foundational transformer for sleep (PFTSleep) model has been trained on more than one million hours of sleep, marking it as one of the “largest” studies in this field.
Go deeper with GlobalData
PFTSleep leverages a transformer-based AI model to “analyse” eight-hour sleep signals, including brain waves, heart rate, breathing patterns, and activity of the muscles.
This approach enables the tool to generate detailed summaries and categorise sleep stages for the complete night.
Mount Sinai states that current methods of sleep analysis often depend on manual scoring by human experts or AI models that cannot analyse a full night’s sleep.
The PFTSleep model, however, utilises a vast dataset of sleep studies and allows for a “standardised” and “scalable” method for research in sleep and clinical applications, stated the investigators.
The model is trained through self-supervision, a method that aids in learning clinical features from physiological signals without relying on human-labelled results.
Across various settings, the model claims to detect patterns of sleep during the night.
Icahn School of Medicine’s AI and emerging technologies training area PhD candidate Benjamin Fox said: “This is a step forward in AI-assisted sleep analysis and interpretation.
“By leveraging AI in this way, we can learn relevant clinical features directly from sleep study signal data and use them for sleep scoring and, in the future, other clinical applications such as detecting sleep apnoea or assessing health risks linked to sleep quality.”
While the AI tool represents a significant advancement, the researchers underline that it is intended to support, not replace, the expertise of sleep specialists.
The next phase of research will focus on broadening the capabilities of the model to include identifying sleep disorders and predicting outcomes of health.
