Apnea-Hypopnea Index Prediction for Obstructive Sleep Apnea Using Unsegmented SpO2 Signals and Deep Learning

This paper presents an apnea-hypopnea index (AHI) prediction model for obstructive sleep apnea (OSA) by using unsegmented peripheral oxygen saturation (SpO(2)) signals. This proposal, directly predicting AHI values with respect to overnight unsegmented SpO(2) signals, is the first report in the literature, and simply solves the method limitation of our previous study. As well, this approach can provide more features of OSA assessment to users and doctors. Experimental results show that the presented model gives an overall accuracy up to 81.13% for four-level OSA severity classification, which is higher than our original work and significantly outperforms most counterparts in the literature. This work can be used as an easy-to-use and effective screening tool for OSA before undergoing polysomnography (PSG). Moreover, doctors can arrange timely PSG tests for those who require preferential medical care according to the predicted OSA severity.