Abstract:Objective To compare the differences in hypoxia parameters between rapid eye movement（REM）and non - rapid eye movement（NREM）sleep stages in patients with obstructive sleep apnea（OSA）and to construct an artificial neural network（ANN）sleep staging prediction model. Methods A retrospective analysis was performed for86 adult patients who underwent overnight polysomnography（PSG），and exported PSG data files to Matlab software for analysis，and the REM staging events（2 023）and NREM staging events（10 075）with decreased pulse oxygensaturation（SpO2）were extracted. The differences of hypoxia parameters between REM and NREM were analyzed. The ANN model was constructed using a feed-forward structure incorporating a multilayer perceptron（MLP）with a back-propagation algorithm. Predictive performance was assessed using receiver operating characteristic（ROC）curves.Results Compared with the hypoxia parameters in NREM stage，e-minSpO2 and r.DSpO2 were lower，and ΔSpO2，d.DSpO2，ODR，ORR，T90，d.T90，r.T90，and ST90，d.ST90，r.ST90 were higher in REM stage（P<0.05）. The accuracy of ANN model test sets in predicting REM sleep was 84.00%，with the area under the curve（AUC）of 0.73，and the sensitivity，specificity，positive predictive value，and negative predictive value were 0.11，0.99，0.65 and 0.85，respectively. Conclusion There are differences in REM and NREM sleep hypoxia parameters in OSA patients，based on which an ANN prediction model can be constructed to achieve convenient，accurate and rapid identification ofsleep staging，which can provide a reference for the diagnosis and treatment of clinical sleep-related diseases.