Detrended Fluctuation Analysis of Oxyhemoglobin Saturation by Pulse Oximetry in Sleep Apnea Syndrome

Detrended fluctuation analysis of the oximetry signal to assist in paediatric sleep apnoea–hypopnoea syndrome diagnosis

“…The DFA profile shows changes in the correlation properties for different ranges of scales, termed 'crossovers', which may be caused by different non-stationarities in the signal such as (Chen et al 2002): (i) segments removed from the signal; (ii) random spikes with variable amplitude; (iii) segments with different local behaviour. Segments of the SpO 2 signal associated with apnoeic events typically have different statistical properties, presenting fluctuations and spikes (Crespo et al 2017, Hua andYu 2017). Thus, these properties of the SpO 2 signal may be reflected in the DFA profile.…”

“…A double logarithmic plot was used to analyse the evolution of the DFA plot along scales: log (F(k)) versus log (k) (Penzel et al 2003, Dehkordi et al 2016, Hua and Yu 2017. Figure 1 shows the averaged DFA plot for the four SAHS severity groups (AHI < 1 e h −1 , 1 ⩽ AHI < 5 e h −1 , 5 ⩽ AHI < 10 e h −1 , and AHI ⩾ 10 e h −1 ) in the training set.…”

“…A crossover is produced in the space between these two regions of the DFA profile. Robust linear regression (Hua and Yu 2017) was applied to estimate the line that fits both regions for each SpO 2 recording.…”

“…Figure 2 shows the lines fitted in both regions in an illustrative example of a patient from the training set. We characterised the DFA plot by extracting the following features, as can be seen in figure 2 (Penzel et al 2003, Hua andYu 2017):…”

“…Artefacts were rejected from oximetric recordings by removing those SpO 2 values below 50% and sudden changes between consecutive SpO 2 samples faster than 4%/second (Magalang et al 2003). Then, a non-overlapping averaging-window of 1 s was applied (effective sampling rate = 1 Hz) to speed up the signal processing stage, which has been found to be appropriate to perform a multiscale analysis of the oximetry signal (Crespo et al 2017, Hua andYu 2017). This window size is lower than 3 s, which is the maximum averaging-time recommended by the American Academy of Sleep Medicine (AASM) (Berry et al 2012).…”