Severity Prediction of Obstructive Sleep Apnea Using Transformed 2D Oxygen Saturation Signals

SpO 2 ), deep learningIn this paper, we present a deep-learning-based method for predicting obstructive sleep apnea (OSA) severity using 2D oxygen saturation (SpO 2 ) signal representations.Our previous works have demonstrated that unsegmented overnight 1D SpO 2 signals can achieve high accuracy in predicting the apnea-hypopnea index and classifying OSA severity.Extending this approach, we introduce a novel transformation of the original 1D SpO 2 signal into a 2D representation, enabling it to be processed similarly to an image signal and then trained by deep neural networks.This signal transformation is designed to facilitate subsequent studies into model interpretability.Experimental results show that the proposed approach reaches competitive performance compared with state-of-the-art baselines.