The classification of oximetry signals using Bayesian neural networks to assist in the detection of obstructive sleep apnoea syndrome

A Review of Novel Oximetry Parameters for the Prediction of Cardiovascular Disease in Obstructive Sleep Apnoea

Obstructive sleep apnoea (OSA) is a sleep disorder with repetitive collapse of the upper airway during sleep, which leads to intermittent hypoxic events overnight, adverse neurocognitive, metabolic complications, and ultimately an increased risk of cardiovascular disease (CVD). The standard diagnostic parameter for OSA, apnoea-hypopnoea index (AHI), is inadequate to predict CVD morbidity and mortality, because it focuses only on the frequency of apnoea and hypopnoea events, and fails to reveal other physiological information for the prediction of CVD events. Novel parameters have been introduced to compensate for the deficiencies of AHI. However, the calculation methods and criteria for these parameters are unclear, hindering their use in cross-study analysis and studies. This review aims to discuss novel parameters for predicting CVD events from oximetry signals and to summarise the corresponding computational methods.

Application of artificial intelligence in the diagnosis of sleep apnea

STUDY OBJECTIVES

Machine learning (ML) models have been employed in the setting of sleep disorders. This review aims to summarize the existing data about the role of ML techniques in the diagnosis, classification, and treatment of sleep-related breathing disorders.

METHODS

A systematic search in Medline, EMBASE, and Cochrane databases through January 2022 was performed.

RESULTS

Our search strategy revealed 132 studies that were included in the systematic review. Existing data show that ML models have been successfully used for diagnostic purposes. Specifically, ML models showed good performance in diagnosing sleep apnea using easily obtained features from the electrocardiogram, pulse oximetry, and sound signals. Similarly, ML showed good performance for the classification of sleep apnea into obstructive and central categories, as well as predicting apnea severity. Existing data show promising results for the ML-based guided treatment of sleep apnea. Specifically, the prediction of outcomes following surgical treatment and optimization of continuous positive airway pressure therapy can be guided by ML models.

CONCLUSIONS

The adoption and implementation of ML in the field of sleep-related breathing disorders is promising. Advancements in wearable sensor technology and ML models can help clinicians predict, diagnose, and classify sleep apnea more accurately and efficiently.

CITATION

Bazoukis G, Bollepalli SC, Chung CT, et al. Application of artificial intelligence in the diagnosis of sleep apnea. J Clin Sleep Med. 2023;19(7):1337-1363.

Combining Heart Rate Variability and Oximetry to Improve Apneic Event Screening in Non-Desaturating Patients

In this paper, we thoroughly analyze the detection of sleep apnea events in the context of Obstructive Sleep Apnea (OSA), which is considered a public health problem because of its high prevalence and serious health implications. We especially evaluate patients who do not always show desaturations during apneic episodes (non-desaturating patients). For this purpose, we use a database (HuGCDN2014-OXI) that includes desaturating and non-desaturating patients, and we use the widely used Physionet Apnea Dataset for a meaningful comparison with prior work. Our system combines features extracted from the Heart-Rate Variability (HRV) and SpO₂, and it explores their potential to characterize desaturating and non-desaturating events. The HRV-based features include spectral, cepstral, and nonlinear information (Detrended Fluctuation Analysis (DFA) and Recurrence Quantification Analysis (RQA)). SpO₂-based features include temporal (variance) and spectral information. The features feed a Linear Discriminant Analysis (LDA) classifier. The goal is to evaluate the effect of using these features either individually or in combination, especially in non-desaturating patients. The main results for the detection of apneic events are: (a) Physionet success rate of 96.19%, sensitivity of 95.74% and specificity of 95.25% (Area Under Curve (AUC): 0.99); (b) HuGCDN2014-OXI of 87.32%, 83.81% and 88.55% (AUC: 0.934), respectively. The best results for the global diagnosis of OSA patients (HuGCDN2014-OXI) are: success rate of 95.74%, sensitivity of 100%, and specificity of 89.47%. We conclude that combining both features is the most accurate option, especially when there are non-desaturating patterns among the recordings under study.

A novel, simple, and accurate pulse oximetry indicator for screening adult obstructive sleep apnea

OBJECTIVE

The objective of the study was to develop a multiparametric oximetry indicator (IMp-SpO2) to diagnose obstructive sleep apnea in adults.

MATERIAL AND METHOD

This was an observational, retrospective study of diagnostic accuracy. We included adults who had had a diagnostic polysomnography with few artifacts and a total sleep time of at least 180 min in the sleep laboratory. Obstructive sleep apnea (OSA) was defined as an apnea-hypopnea index (AHI) ≥ 5. The database was randomly divided into an experimental (Exp-G) and validation (Val-G) group. The program calculated several parameters of oxygen saturation variability (Par-VarSpO2): (a) oxygen desaturation index (ODI ≥ 3, 4%) and (b) 90, 95, and 97.5 percentiles of both the number of oxygen desaturations ≥ 3 and 4% (P90-97.5 OD3/4 W5-60) and SpO2 standard deviations in moving windows from 5 to 60 min (P90-P97.5 SDSpO2 W5-10). Area under the ROC curve (AUC-ROC), sensitivity, specificity, positive/negative likelihood ratios, and accuracy were calculated.

RESULTS

Of 1141 adults included in the study, experimental (571) and validation group (570) were similar (women 47% vs 45%, BMI 27.5 kg/m² vs 27.2 kg/m², and AHI 11.7 vs 12, p NS). The IMp-SpO2 developed in the experimental group consisted of a combination of 10 parameters of oxygen saturation variability. The presence of at least one IMp-SpO2 variable had a high diagnostic performance for OSA (sensitivity/specificity/accuracy: Exp-G: 92.8/94/93.2%; Val-G: 93/95.2/93.7%). The IMp-SpO2 AUC-ROC was higher (Exp-G 0.934, Val-G 0.941) than most of the Par-VarSpO2 (0.898-0.929, p < 0.05).

CONCLUSION

The IMp-SpO2 showed a > 90% accuracy for OSA diagnosis in adults.

Comparative associations of oximetry patterns in Obstructive Sleep Apnea with incident cardiovascular disease

STUDY OBJECTIVES

Intermittent hypoxia is a key mechanism linking Obstructive Sleep Apnea (OSA) to cardiovascular disease (CVD). Oximetry analysis could enhance understanding of which OSA phenotypes are associated with CVD risk. The aim of this study was to compare associations of different oximetry patterns with incident CVD in men and women with OSA.

METHODS

Sleep Heart Health Study data were used for analysis. n = 2878 Participants (51.8% female; mean age 63.5 ± 10.5 years) with OSA (Apnea Hypopnea Index [AHI] ≥ 5 events/h) and no pre-existing CVD at baseline or within the first 2 years of follow-up were included. Four oximetry analysis approaches were applied: desaturation characteristics, time series analysis, power spectral density, and non-linear analysis. Thirty-one resulting oximetry patterns were compared to incident CVD using proportional hazards regression models adjusted for age, race, smoking, BMI, and sex.

RESULTS

There were no associations between OSA oximetry patterns and incident CVD in the total sample or in men. In women, there were some associations between incident CVD and time series analysis (e.g. SpO2 distribution standard deviation, HR 0.81, 95% CI 0.68-0.96, p = 0.014) and power spectral density oximetry patterns (e.g. Full frequency band mean HR 0.75; 95% CI 0.59-0.95; p = 0.015).

CONCLUSIONS

Comprehensive comparison of baseline oximetry patterns in OSA found none were related to development of CVD. There were no standout individual oximetry patterns that appear to be candidates for CVD risk phenotyping in OSA, but some showed marginal relationships with CVD risk in women. Further work is required to understand whether OSA phenotypes can be used to predict susceptibility to cardiovascular disease.

Predictive performances of 6 data mining techniques for obstructive sleep apnea-hypopnea syndrome

PURPOSE

This study compared the effects of 6 types of obstructive sleep apnea-hypopnea syndrome (OSAHS) prediction models to develop a reference for selecting OSAHS data mining tools in clinical practice.

METHODS

This cross-sectional study included 401 cases. They were randomly divided into 2 groups: training (70%) and testing (30%). Logistic regression, a Bayesian network, an artificial neural network, a support vector learning machine, C5.0, and a classification and regression tree were each adopted to establish 6 prediction models. After training, the 6 models were used to test the remaining samples and calculate the correct and error rates of each model.

RESULTS

Twenty-one input variables for which the difference between the patient and nonpatient groups was statistically significant were considered. The models found the abdominal circumference, neck circumference, and nocturia ≥2 per night to be the most important variables. The support vector machine, neural network, and C5.0 models performed better than the classification and regression tree, Bayesian network, and logistic regression models.

CONCLUSIONS

In terms of predicting the risk of OSAHS, the support vector machine, neural network, and C5.0 were superior to the classification and regression tree, Bayesian network, and logistic regression models. However, such results were obtained based on the data of a single center, so they need to be further validated by other institutions.

Oximetry Indices in the Management of Sleep Apnea: From Overnight Minimum Saturation to the Novel Hypoxemia Measures

Obstructive sleep apnea (OSA) is a multidimensional disease often underdiagnosed due to the complexity and unavailability of its standard diagnostic method: the polysomnography. Among the alternative abbreviated tests searching for a compromise between simplicity and accurateness, oximetry is probably the most popular. The blood oxygen saturation (SpO₂) signal is characterized by a near-constant profile in healthy subjects breathing normally, while marked drops (desaturations) are linked to respiratory events. Parameterization of the desaturations has led to a great number of indices of severity assessment commonly used to assist in OSA diagnosis. In this chapter, the main methodologies used to characterize the overnight oximetry profile are reviewed, from visual inspection and simple statistics to complex measures involving signal processing and pattern recognition techniques. We focus on the individual performance of each approach, but also on the complementarity among the great amount of indices existing in the state of the art, looking for the most relevant oximetric feature subset. Finally, a quick overview of SpO₂-based deep learning applications for OSA management is carried out, where the raw oximetry signal is analyzed without previous parameterization. Our research allows us to conclude that all the methodologies (conventional, time, frequency, nonlinear, and hypoxemia-based) demonstrate high ability to provide relevant oximetric indices, but only a reduced set provide non-redundant complementary information leading to a significant performance increase. Finally, although oximetry is a robust tool, greater standardization and prospective validation of the measures derived from complex signal processing techniques are still needed to homogenize interpretation and increase generalizability.

Wavelet Analysis of Overnight Airflow to Detect Obstructive Sleep Apnea in Children

This study focused on the automatic analysis of the airflow signal (AF) to aid in the diagnosis of pediatric obstructive sleep apnea (OSA). Thus, our aims were: (i) to characterize the overnight AF characteristics using discrete wavelet transform (DWT) approach, (ii) to evaluate its diagnostic utility, and (iii) to assess its complementarity with the 3% oxygen desaturation index (ODI3). In order to reach these goals, we analyzed 946 overnight pediatric AF recordings in three stages: (i) DWT-derived feature extraction, (ii) feature selection, and (iii) pattern recognition. AF recordings from OSA patients showed both lower detail coefficients and decreased activity associated with the normal breathing band. Wavelet analysis also revealed that OSA disturbed the frequency and energy distribution of the AF signal, increasing its irregularity. Moreover, the information obtained from the wavelet analysis was complementary to ODI3. In this regard, the combination of both wavelet information and ODI3 achieved high diagnostic accuracy using the common OSA-positive cutoffs: 77.97%, 81.91%, and 90.99% (AdaBoost.M2), and 81.96%, 82.14%, and 90.69% (Bayesian multi-layer perceptron) for 1, 5, and 10 apneic events/hour, respectively. Hence, these findings suggest that DWT properly characterizes OSA-related severity as embedded in nocturnal AF, and could simplify the diagnosis of pediatric OSA.

Usefulness of recurrence plots from airflow recordings to aid in paediatric sleep apnoea diagnosis

BACKGROUND AND OBJECTIVE

In-laboratory overnight polysomnography (PSG) is the gold standard method to diagnose the Sleep Apnoea-Hypopnoea Syndrome (SAHS). PSG is a complex, expensive, labour-intensive and time-consuming test. Consequently, simplified diagnostic methods are desirable. We propose the analysis of the airflow (AF) signal by means of recurrence plots (RP) features. The main goal of our study was to evaluate the utility of the information from RPs of the AF signals to detect paediatric SAHS at different levels of severity. In addition, we also evaluated the complementarity with the 3% oxygen desaturation index (ODI₃).

METHODS

946 AF and blood oxygen saturation (SpO₂) recordings from children ages 0-13 years were used. The population under study was randomly split into training (60%) and test (40%) sets. RP was computed and 9 RP features were extracted from each AF recording. ODI₃ was also calculated from each SpO₂ recording. A feature selection stage was conducted in the training group by means of the fast correlation-based filter (FCBF) methodology to obtain a relevant and non-redundant optimum feature subset. A multi-layer perceptron neural network with Bayesian approach (BY-MLP), trained with these optimum features, was used to estimate the apnoea-hypopnoea index (AHI).

RESULTS

8 of the RP features showed statistically significant differences (p-value <0.01) among the SAHS severity groups. FCBF selected the maximum length of the diagonal lines from RP, as well as the ODI₃. Using these optimum features, the BY-MLP model achieved 83.2%, 78.5%, and 91.0% accuracy in the test group for the AHI thresholds 1, 5, and 10 events/h, respectively. Moreover, this model reached a negative likelihood ratio of 0.1 for 1 event/h and a positive likelihood ratio of 13.7 for 10 events/h.

CONCLUSIONS

RP analysis enables extraction of useful SAHS-related information from overnight AF paediatric recordings. Moreover, it provides complementary information to the widely-used clinical variable ODI₃. Thus, RP applied to AF signals can be used along with ODI₃ to help in paediatric SAHS diagnosis, particularly to either confirm the absence of SAHS or the presence of severe SAHS.

Bispectral Analysis to Enhance Oximetry as a Simplified Alternative for Pediatric Sleep Apnea Diagnosis

This study aims at assessing the bispectral analysis of blood oxygen saturation (SpO²) from nocturnal oximetry to help in pediatric sleep apnea-hypopnea syndrome (SAHS) diagnosis. Recent studies have found excessive redundancy in the SAHS-related information usually extracted from SpO², while proposing only two features as a reduced set to be used. On the other hand, it has been suggested that SpO₂ bispectral analysis is able to provide complementary information to common anthropometric, spectral, and clinical variables. We address these novel findings to assess whether bispectrum provides new non-redundant information to help in SAHS diagnosis. Thus, we use 981 pediatric SpO₂ recordings to extract both the reduced set of features recently proposed as well as 9 bispectral features. Then, a feature selection method based on the fast correlationbased filter and bootstrapping is used to assess redundancy among all the features. Finally, the non-redundant ones are used to train a Bayesian multi-layer perceptron neural network (BYMLP) that estimate the apnea-hypopnea index (AHI), which is the diagnostic reference variable. Bispectral phase entropy was found complementary to the two previously recommended features and a BY-MLP model trained with the three of them reached high agreement with actual AHI (intra-class correlation coefficient = 0.889). Estimated AHI also showed high diagnostic ability, reaching 82.1%, 81.9%, and 90.3% accuracies and 0.814, 0.880, and 0.922 area under the receiver-operating characteristics curve for three common AHI thresholds: 1 e/h, 5 e/h, and 10 e/h, respectively. These results suggest that the information extracted from the bispectrum of SpO₂ can improve the diagnostic performance of the oximetry test.

A Bayesian neural network approach to compare the spectral information from nasal pressure and thermistor airflow in the automatic sleep apnea severity estimation

In the sleep apnea-hypopnea syndrome (SAHS) context, airflow signal plays a key role for the simplification of the diagnostic process. It is measured during the standard diagnostic test by the acquisition of two simultaneous sensors: a nasal prong pressure (NPP) and a thermistor (TH). The current study focuses on the comparison of their spectral content to help in the automatic SAHS-severity estimation. The spectral analysis of 315 NPP and corresponding TH recordings is firstly proposed to characterize the conventional band of interest for SAHS (0.025-0.050 Hz.). A magnitude squared coherence analysis is also conducted to quantify possible differences in the frequency components of airflow from both sensors. Then, a feature selection stage is implemented to assess the relevance and redundancy of the information extracted from the spectrum of NPP and TH airflow. Finally, a multiclass Bayesian multi-layer perceptron (BY-MLP) was used to perform an automatic estimation of SAHS severity (no-SAHS, mild, moderate, and severe), by the use of the selected spectral features from: airflow NPP alone, airflow TH alone, and both sensors jointly. The highest diagnostic performance was reached by BY-MLP only trained with NPP spectral features, reaching Cohen's κ = 0.498 in the overall four-class classification task. It also achieved 91.3%, 84.9%, and 83.3% of accuracy in the binary evaluation of the 3 apnea-hypopnea index cut-offs (5, 15, and 30 events/hour) that define the four SAHS degrees. Our results suggest that TH sensor might be not necessary for SAHS severity estimation if an automatic comprehensive characterization approach is adopted to simplify the diagnostic process.

Predictive monitoring of mobile patients by combining clinical observations with data from wearable sensors

The majority of patients in the hospital are ambulatory and would benefit significantly from predictive and personalized monitoring systems. Such patients are well suited to having their physiological condition monitored using low-power, minimally intrusive wearable sensors. Despite data-collection systems now being manufactured commercially, allowing physiological data to be acquired from mobile patients, little work has been undertaken on the use of the resultant data in a principled manner for robust patient care, including predictive monitoring. Most current devices generate so many false-positive alerts that devices cannot be used for routine clinical practice. This paper explores principled machine learning approaches to interpreting large quantities of continuously acquired, multivariate physiological data, using wearable patient monitors, where the goal is to provide early warning of serious physiological determination, such that a degree of predictive care may be provided. We adopt a one-class support vector machine formulation, proposing a formulation for determining the free parameters of the model using partial area under the ROC curve, a method arising from the unique requirements of performing online analysis with data from patient-worn sensors. There are few clinical evaluations of machine learning techniques in the literature, so we present results from a study at the Oxford University Hospitals NHS Trust devised to investigate the large-scale clinical use of patient-worn sensors for predictive monitoring in a ward with a high incidence of patient mortality. We show that our system can combine routine manual observations made by clinical staff with the continuous data acquired from wearable sensors. Practical considerations and recommendations based on our experiences of this clinical study are discussed, in the context of a framework for personalized monitoring.

Gaussian processes for personalized e-health monitoring with wearable sensors

Advances in wearable sensing and communications infrastructure have allowed the widespread development of prototype medical devices for patient monitoring. However, such devices have not penetrated into clinical practice, primarily due to a lack of research into "intelligent" analysis methods that are sufficiently robust to support large-scale deployment. Existing systems are typically plagued by large false-alarm rates, and an inability to cope with sensor artifact in a principled manner. This paper has two aims: 1) proposal of a novel, patient-personalized system for analysis and inference in the presence of data uncertainty, typically caused by sensor artifact and data incompleteness; 2) demonstration of the method using a large-scale clinical study in which 200 patients have been monitored using the proposed system. This latter provides much-needed evidence that personalized e-health monitoring is feasible within an actual clinical environment, at scale, and that the method is capable of improving patient outcomes via personalized healthcare.

Gaussian process regression in vital-sign early warning systems

The current standard of clinical practice for patient monitoring in most developed nations is connection of patients to vital-sign monitors, combined with frequent manual observation. In some nations, such as the UK, manual early warning score (EWS) systems have been mandated for use, in which scores are assigned to the manual observations, and care escalated if the scores exceed some pre-defined threshold. We argue that this manual system is far from ideal, and can be improved using machine learning techniques. We propose a system based on Gaussian process regression for improving the efficacy of existing EWS systems, and then demonstrate the method using manual observation of vital signs from a large-scale clinical study.

Feature selection from nocturnal oximetry using genetic algorithms to assist in obstructive sleep apnoea diagnosis

Nocturnal pulse oximetry (NPO) has demonstrated to be a powerful tool to help in obstructive sleep apnoea (OSA) detection. However, additional analysis is needed to use NPO alone as an alternative to nocturnal polysomnography (NPSG), which is the gold standard for a definitive diagnosis. In the present study, we exhaustively analysed a database of blood oxygen saturation (SpO(2)) recordings (80 OSA-negative and 160 OSA-positive) to obtain further knowledge on the usefulness of NPO. Population set was randomly divided into training and test sets. A feature extraction stage was carried out: 16 features (time and frequency statistics and spectral and nonlinear features) were computed. A genetic algorithm (GA) approach was applied in the feature selection stage. Our methodology achieved 87.5% accuracy (90.6% sensitivity and 81.3% specificity) in the test set using a logistic regression (LR) classifier with a reduced number of complementary features (3 time domain statistics, 1 frequency domain statistic, 1 conventional spectral feature and 1 nonlinear feature) automatically selected by means of GAs. Our results improved diagnostic performance achieved with conventional oximetric indexes commonly used by physicians. We concluded that GAs could be an effective and robust tool to search for essential oximetric features that could enhance NPO in the context of OSA diagnosis.

Automatic detection of apnoea of prematurity

The detection of the incidents of apnoea of prematurity (AP) in preterm infants is important in the intensive care unit, but this detection is often based on simple threshold techniques, which suffer from poor specificity. Three methods for the automatic detection of AP were designed, tested and evaluated using approximately 2426 h of continuous recording from 54 neonates (μ = 44 h and σ = 7 h). The first method was based on the cumulative sum of the time series of heart rate (HR), respiratory rate (RR) and oxygen saturation (SpO(2)) along with the sum of their Shannon entropy. The performance of this method gave 94.53% sensitivity, 74.72% specificity and 77.84% accuracy. The second method was based on the correlation between the time series of HR, RR and SpO(2), which were used as inputs to an artificial neural network. This gave 81.85% sensitivity, 75.83% specificity and 76.78% accuracy. The third method utilized the derivative of the three time series and yielded a performance of 100% sensitivity, 96.19% specificity and 96.79% accuracy. Although not optimized to work in real time, the latter method has the potential for forming the basis of a real time system for the detection of incidents of AP.