Intra-subject variability of snoring sounds in relation to body position, sleep stage, and blood oxygen level.

摘要

In a multidimensional feature space, the snoring sounds can extend from a very compact cluster to highly distinct clusters. In this study, we investigated the cause of snoring sound's variation within the snorers. It is known that a change in body position and sleep stage can affect snoring during sleep but it is unclear whether positional, sleep state, and blood oxygen level variations cause the snoring sounds to have different characteristics, and if it does how significant that effect would be. We extracted 12 characteristic features from snoring sound segments of 57 snorers and transformed them into a 4-D feature space using principal component analysis (PCA). Then, they were grouped based on the body position (side, supine, and prone), sleep stage (NREM, REM, and Arousal), and blood oxygen level (Normal and Desaturation). The probability density function of the transformed features was calculated for each class of categorical variables. The distance between the class-densities were calculated to determine which of these parameters affects the snoring sounds significantly. Analysis of Variance (ANOVA) was run for each categorical variable. The results show that the positional change has the highest effect on the snoring sounds; it results in forming distinct clusters of snoring sounds. Also, sleep state and blood oxygen level variation have been found to moderately affect the snoring sounds.