0440 Validation Study of Neural Network Algorithm for Automated Sleep Stage Scoring: StageNet

• Published in: Sleep (New York, N.Y.) Vol. 43; no. Supplement_1; p. A169
• Main Authors: Choi, J, Moon, J
• Format: Journal Article
• Language: English
• Published: US Oxford University Press 27.05.2020
• Subjects: Algorithms; Automation; Datasets; Neural networks; Sleep; Validation studies
• ISSN: 0161-8105; 1550-9109; 1550-9109
• DOI: 10.1093/sleep/zsaa056.437

Abstract Introduction Polysomnography (level 1) is an important test for evaluating sleep status or disorder. In general, measured raw data of polysomnography is manually scored by sleep expert. However, manual scoring is a time-consuming and labor-intensive work. The purpose of this study was to verify the accuracy of automated sleep stage scoring based on the neural network algorithm compared to the manual sleep stage scoring. Methods A total 604 polysomnography data set of subjects (Male: Female = 409: 195) aged 19 to 60 years were finally included in the study. The performance of proposed model was evaluated with kappa and bootstrapped point-estimated of median percent agreement with 95% percentile bootstrap confidence interval with R=1000. The proposed model is trained with 484 data set and validated with 48 data set. For test, 72 data set are randomly selected. Results The proposed model showed good concordance rates in stage W (94%), N1 (83.9%), N2 (89%), N3 (92%) and R (93%) between automated neural network algorithm and manual scoring. The average kappa value was 0.85. In bootstrap method, high overall agreements of automated neural network algorithm were found in stage W (98%), N1 (94%), N2 (92%), N3 (99%), R (98%) and total (96%). Conclusion Automated sleep stage scoring using proposed model - StageNet may be a reliable method for sleep stage classification. Support