Automatic sleep staging using a small-footprint sensor array and recurrent-convolutional neural networks

The accelerating trend towards personalized 'pre-cision medicine' and tele-healthcare is revolutionizing the practice of medicine and giving the individual unprecedented access to their own health data. At the same time, a widening gap between wakeful health (ex. physical activity) and nocturnal health (sleep) has revealed the need for accurate, reliable and automated methods to measure sleep in the home. Here we describe a small-footprint sensor array, using electrode stickers that can be self-applied to the forehead, in conjunction with an automated scoring algorithm that achieves accuracies on par with trained human experts (77% agreement using a five-class taxonomy). Compared to alternatives, this approach avoids the low signal-to-noise ratios of dry-contact scalp electrodes while also circumventing the need to measure through hair. Critically, it does not require a trained human expert, either to apply the electrodes or to translate the signals into a useful description of sleep patterns. Taken together, this represents an exciting step forward towards affordable, reliable, and accurate in-the-home sleep assessment.