Figure 1.
A mobile device that integrates a pulse oximeter with a smartphone.
Table 1.
Demographic and PSG information in the study group (mean ± standard deviation).
Figure 2.
Power spectral density applied to a 2-minute SpO2 signal of (A) a child with, and (B) without SDB.
The SDB child shows a clear modulation frequency peak, whereas the NonSDB child illustrates no clear modulation peak.
Figure 3.
PSD applied to the SpO2 of whole study population.
The mean PSD (average of the PSDs obtained for each time window overnight) for each SDB subject is represented in light grey, and NonSDB is represented in dark grey.
Figure 4.
Time-varying power spectral density applied to an overnight SpO2 signal of (A) a child with and (B) without SDB.
The SDB child shows a clear modulation frequency peak and higher energy around this peak compared to the NonSDB child.
Table 2.
Parameter description and corresponding statistics.
Figure 5.
Diagram of the classification process with internal LOO and external 4-fold cross-validation (CV).
The dataset was randomly divided into 4 non-overlapping subsets. 3 formed the training dataset and the remaining formed the test dataset. This process was repeated four times, until each subset was treated once as the test dataset. The most discriminant features classifying children with and without SD were selected using LOO-CV. This feature selection was then evaluated with the independent test dataset.
Figure 6.
Distribution of SpO2 pattern characterization features.
Boxplot of some features extracted from SpO2 pattern characterization such as (A) the mean of Δ (M_Δ), (B) the number of desaturations of 2% below baseline (M_n2%), (C) the spectral power in the modulation band (M_P), and (D) the spectral Shannon entropy (M_SE). Children with SDB show higher SpO2 variability reflected M_Δ and a higher number of desaturations M_n2% due to sleep apnea. They also reflect higher power in the modulation band and lower spectral complexity (see Table 3). Quartile values are displayed as bottom, middle and top horizontal line of the boxes. Whiskers are used to represent the most extreme values within 1.5 times the interquartile range from the median. Outliers (data with values beyond the ends of the whiskers) are displayed as circles.
Table 3.
Effects of SDB on the normally distributed features extracted from SpO2 pattern characterization and PRV analysis.
Table 4.
Effects of SDB on the non-normally distributed features extracted from SpO2 pattern characterization and PRV analysis.
Figure 7.
Boxplot of features extracted from PRV analysis such as (A) the mean of pulse to pulse intervals (M_RR), (B) the standard deviation of RMSSD (S_RMSSD) in time domain, (C) the standard deviation of the normalized power in the LF, and (D) HF band in the spectral domain (S_LF and S_HF, respectively). Children with SDB reflect higher heart rate and PRV dispersion, reflected by a lower pulse to pulse interval and higher standard deviation of the standard PRV measures. Quartile values are displayed as bottom, middle and top horizontal lines of the boxes. Whiskers are used to represent the most extreme values within 1.5 times the interquartile range from the median. Outliers (data with values beyond the ends of the whiskers) are displayed as circles.
Figure 8.
Performance of the feature selection.
The performance is represented in terms of accuracy, sensitivity and specificity in classifying SDB and NonSDB children, whenever the feature that provided the higher AUC (with the training set) was included in the linear discriminant. The results obtained with (A) the training dataset (with internal LOO cross-validation) and (B) the test dataset (with external 4-fold cross-validation) are illustrated.
Figure 9.
The ROC obtained with the 8 most discriminating features (see Table 5) applied to (A) the training dataset (with internal LOO cross-validation) and (B) the test dataset (with external 4-fold cross-validation).
Table 5.
Classification performance based on the linear discriminant analysis using the most discriminatory set of 8 features.
Figure 10.
The histogram of the feature selection process, where the 15 most discriminating features were selected in each iteration. The histogram illustrates the total number of times each feature was automatically chosen by the selection algorithm in each iteration (4 in total). The feature selection was validated internally with a LOO cross-validation and externally with 4-fold cross-validation. The features selected in every iteration (4 times) or nearly every iteration (3 times) were defined as the most discriminating and were proposed as the optimal to create the final linear discriminant. They are represented in black and marked with *.
Figure 11.
SpO2 signal with different resolutions.
An SpO2 signal segment, recorded using the Phone Oximeter (0.1% resolution) for (A) SDB and (B) NonSDB children, and the corresponding SpO2 signal recorded simultaneously with the PSG's pulse oximeter (1% resolution) for the same SDB (C) and NonSDB (D) children. The SpO2 resolution has a great influence in regularity measures like approximate entropy and Lempel-Ziv [41]. Therefore, SpO2 resolution should be taken into account when studying the SpO2 pattern in children with SDB.The SpO2 randomness shown for NonSDB children in 0.1% resolution SpO2 signal (provided by the Phone Oximeter), is not reflected in the 1% resolution SpO2 signal (provided by the PSG's pulse oximeter) because of the rounding effect. This resolution difference might be the reason why children with SDB showed a higher complexity than NonSDB children with conventional pulse oximeter.