According to a recent study, up to one billion people worldwide suffer from obstructive sleep apnea (OSA), a nocturnal breathing disorder that causes a major impact on healthcare systems and national economies.
Helping physicians more effectively identify sleep stages is essential in the diagnostics of sleep disorders including OSA. A team of researchers in Finland developed a deep learning model that automatically recognizes sleep stages as well as an experienced physician.
Published on the IEEE Journal of Biomedical and Health Informatics late last year, the research for Accurate Deep Learning-Based Sleep Staging in a Clinical Population with Suspected Obstructive Sleep Apnea was performed by researchers from the Kuopio University Hospital, and the University of Eastern Finland.
“Identification of sleep stages is crucial in diagnostics of various sleep disorders,” the scientists said. “We aimed to develop an accurate deep learning-based automatic method for the classification of sleep stages in patients with suspected OSA.”
To do this, the team used polysomnographic recording data from healthy individuals, and others suspected of having OSA, to develop a deep learning model for automatic classification of sleep stages.
“The proposed deep learning-based automatic method enables reliable, fast, and accurate sleep staging for suspected OSA patients,” the scientists explained. The accuracy of the sleep staging decreases with increasing OSA severity but with the utilized large clinical dataset, the sleep staging can be conducted for patients suffering from OSA with almost comparable accuracy to individuals without OSA.”
In terms of accuracy, the model, based on a convolutional neural network and recurrent neural network, had a single-channel accuracy for identifying sleep stages of 82.9% on the clinical dataset, and 83.7% on the public data set. The accuracy achieved on the public dataset was superior to previously published state-of-the-art methods.
For training, the team used an NVIDIA RTX 2080Ti GPU, with the cuDNN-accelerated TensorFlow deep learning framework for both training and inference.