A Survey on Recent Advances in Machine Learning Based Sleep Apnea Detection Systems

Towards automatic home-based sleep apnea estimation using deep learning

Apnea and hypopnea are common sleep disorders characterized by the obstruction of the airways. Polysomnography (PSG) is a sleep study typically used to compute the Apnea-Hypopnea Index (AHI), the number of times a person has apnea or certain types of hypopnea per hour of sleep, and diagnose the severity of the sleep disorder. Early detection and treatment of apnea can significantly reduce morbidity and mortality. However, long-term PSG monitoring is unfeasible as it is costly and uncomfortable for patients. To address these issues, we propose a method, named DRIVEN, to estimate AHI at home from wearable devices and detect when apnea, hypopnea, and periods of wakefulness occur throughout the night. The method can therefore assist physicians in diagnosing the severity of apneas. Patients can wear a single sensor or a combination of sensors that can be easily measured at home: abdominal movement, thoracic movement, or pulse oximetry. For example, using only two sensors, DRIVEN correctly classifies 72.4% of all test patients into one of the four AHI classes, with 99.3% either correctly classified or placed one class away from the true one. This is a reasonable trade-off between the model's performance and the patient's comfort. We use publicly available data from three large sleep studies with a total of 14,370 recordings. DRIVEN consists of a combination of deep convolutional neural networks and a light-gradient-boost machine for classification. It can be implemented for automatic estimation of AHI in unsupervised long-term home monitoring systems, reducing costs to healthcare systems and improving patient care.

Algorithmic detection of sleep-disordered breathing using respiratory signals: a systematic review

Background and Objective. Sleep-disordered breathing (SDB) poses health risks linked to hypertension, cardiovascular disease, and diabetes. However, the time-consuming and costly standard diagnostic method, polysomnography (PSG), limits its wide adoption and leads to underdiagnosis. To tackle this, cost-effective algorithms using single-lead signals (like respiratory, blood oxygen, and electrocardiogram) have emerged. Despite respiratory signals being preferred for SDB assessment, a lack of comprehensive reviews addressing their algorithmic scope and performance persists. This paper systematically reviews 2012-2022 literature, covering signal sources, processing, feature extraction, classification, and application, aiming to bridge this gap and provide future research references.Methods. This systematic review followed the registered PROSPERO protocol (CRD42022385130), initially screening 342 papers, with 32 studies meeting data extraction criteria.Results. Respiratory signal sources include nasal airflow (NAF), oronasal airflow (OAF), and respiratory movement-related signals such as thoracic respiratory effort (TRE) and abdominal respiratory effort (ARE). Classification techniques include threshold rule-based methods (8), machine learning models (13), and deep learning models (11). The NAF-based algorithm achieved the highest average accuracy at 94.11%, surpassing 78.19% for other signals. Hypopnea detection sensitivity with single-source respiratory signals remained modest, peaking at 73.34%. The TRE and ARE signals proved to be reliable in identifying different types of SDB because distinct respiratory disorders exhibited different patterns of chest and abdominal motion.Conclusions. Multiple detection algorithms have been widely applied for SDB detection, and their accuracy is closely related to factors such as signal source, signal processing, feature selection, and model selection.

Pediatric obstructive sleep apnea diagnosis: leveraging machine learning with linear discriminant analysis

Objective

The objective of this study was to investigate the effectiveness of a machine learning algorithm in diagnosing OSA in children based on clinical features that can be obtained in nonnocturnal and nonmedical environments.

Patients and methods

This study was conducted at Beijing Children's Hospital from April 2018 to October 2019. The participants in this study were 2464 children aged 3-18 suspected of having OSA who underwent clinical data collection and polysomnography(PSG). Participants' data were randomly divided into a training set and a testing set at a ratio of 8:2. The elastic net algorithm was used for feature selection to simplify the model. Stratified 10-fold cross-validation was repeated five times to ensure the robustness of the results.

Results

Feature selection using Elastic Net resulted in 47 features for AHI ≥5 and 31 features for AHI ≥10 being retained. The machine learning model using these selected features achieved an average AUC of 0.73 for AHI ≥5 and 0.78 for AHI ≥10 when tested externally, outperforming models based on PSG questionnaire features. Linear Discriminant Analysis using the selected features identified OSA with a sensitivity of 44% and specificity of 90%, providing a feasible clinical alternative to PSG for stratifying OSA severity.

Conclusions

This study shows that a machine learning model based on children's clinical features effectively identifies OSA in children. Establishing a machine learning screening model based on the clinical features of the target population may be a feasible clinical alternative to nocturnal OSA sleep diagnosis.

Long-term health outcomes for patients with obstructive sleep apnea: placing the Agency for Healthcare Research and Quality report in context-a multisociety commentary

This multisociety commentary critically examines the Agency for Healthcare Research and Quality (AHRQ) final report and systematic review on long-term health outcomes in obstructive sleep apnea. The AHRQ report was commissioned by the Centers for Medicare & Medicaid Services and particularly focused on the long-term patient-centered outcomes of continuous positive airway pressure, the variability of sleep-disordered breathing metrics, and the validity of these metrics as surrogate outcomes. This commentary raises concerns regarding the AHRQ report conclusions and their potential implications for policy decisions. A major concern expressed in this commentary is that the AHRQ report inadequately acknowledges the benefits of continuous positive airway pressure for several established, long-term clinically important outcomes including excessive sleepiness, motor vehicle accidents, and blood pressure. While acknowledging the limited evidence for the long-term benefits of continuous positive airway pressure treatment, especially cardiovascular outcomes, as summarized by the AHRQ report, this commentary reviews the limitations of recent randomized controlled trials and nonrandomized controlled studies and the challenges of conducting future randomized controlled trials. A research agenda to address these challenges is proposed including study designs that may include both high quality randomized controlled trials and nonrandomized controlled studies. This commentary concludes by highlighting implications for the safety and quality of life for the millions of people living with obstructive sleep apnea if the AHRQ report alone was used by payers to limit coverage for the treatment of obstructive sleep apnea while not considering the totality of available evidence.

CITATION

Patil SP, Billings ME, Bourjeily G, et al. Long-term health outcomes for patients with obstructive sleep apnea: placing the Agency for Healthcare Research and Quality report in context-a multisociety commentary. J Clin Sleep Med. 2024;20(1):135-149.

Software defined radio frequency sensing framework for intelligent monitoring of sleep apnea syndrome

Healthy sleep is vital to all functions in the body. It improves physical and mental health, strengthens resistance against diseases, and develops strong immunity against metabolism and chronic diseases. However, a sleep disorder can cause the inability to sleep well. Sleep apnea syndrome is a critical breathing disorder that occurs during sleeping when breathing stops suddenly and starts when awake, causing sleep disturbance. If it is not treated timely, it can produce loud snoring and drowsiness or causes more acute health problems such as high blood pressure or heart attack. The accepted standard for diagnosing sleep apnea syndrome is full-night polysomnography. However, its limitations include a high cost and inconvenience. This article aims to develop an intelligent monitoring framework for detecting breathing events based on Software Defined Radio Frequency (SDRF) sensing and verify its feasibility for diagnosing sleep apnea syndrome. We extract the wireless channel state information (WCSI) for breathing motion using channel frequency response (CFR) recorded in time at every instant at the receiver. The proposed approach simplifies the receiver structure with the added functionality of communication and sensing together. Initially, simulations are conducted to test the feasibility of the SDRF sensing design for the simulated wireless channel. Then, a real-time experimental setup is developed in a lab environment to address the challenges of the wireless channel. We conducted 100 experiments to collect the dataset of 25 subjects for four breathing patterns. SDRF sensing system accurately detected breathing events during sleep without subject contact. The developed intelligent framework uses machine learning classifiers to classify sleep apnea syndrome and other breathing patterns with an acceptable accuracy of 95.9%. The developed framework aims to build a non-invasive sensing system to diagnose patients conveniently suffering from sleep apnea syndrome. Furthermore, this framework can easily be further extended for E-health applications.

Sleep disorder and apnea events detection framework with high performance using two-tier learning model design

Sleep apnea is defined as a breathing disorder that affects sleep. Early detection of sleep apnea helps doctors to take intervention for patients to prevent sleep apnea. Manually making this determination is a time-consuming and subjectivity problem. Therefore, many different methods based on polysomnography (PSG) have been proposed and applied to detect this disorder. In this study, a unique two-layer method is proposed, in which there are four different deep learning models in the deep neural network (DNN), gated recurrent unit (GRU), recurrent neural network (RNN), RNN-based-long term short term memory (LSTM) architecture in the first layer, and a machine learning-based meta-learner (decision-layer) in the second layer. The strategy of making a preliminary decision in the first layer and verifying/correcting the results in the second layer is adopted. In the training of this architecture, a vector consisting of 23 features consisting of snore, oxygen saturation, arousal and sleep score data is used together with PSG data. A dataset consisting of 50 patients, both children and adults, is prepared. A number of pre-processing and under-sampling applications have been made to eliminate the problem of unbalanced classes. Proposed method has an accuracy of 95.74% and 99.4% in accuracy of apnea detection (apnea, hypopnea and normal) and apnea types detection (central, mixed and obstructive), respectively. Experimental results demonstrate that patient-independent consistent results can be produced with high accuracy. This robust model can be considered as a system that will help in the decisions of sleep clinics where it is expected to detect sleep disorders in detail with high performance.

Obstructive sleep apnea event detection using explainable deep learning models for a portable monitor

Background

Polysomnography (PSG) is the gold standard for detecting obstructive sleep apnea (OSA). However, this technique has many disadvantages when using it outside the hospital or for daily use. Portable monitors (PMs) aim to streamline the OSA detection process through deep learning (DL).

Materials and methods

We studied how to detect OSA events and calculate the apnea-hypopnea index (AHI) by using deep learning models that aim to be implemented on PMs. Several deep learning models are presented after being trained on polysomnography data from the National Sleep Research Resource (NSRR) repository. The best hyperparameters for the DL architecture are presented. In addition, emphasis is focused on model explainability techniques, concretely on Gradient-weighted Class Activation Mapping (Grad-CAM).

Results

The results for the best DL model are presented and analyzed. The interpretability of the DL model is also analyzed by studying the regions of the signals that are most relevant for the model to make the decision. The model that yields the best result is a one-dimensional convolutional neural network (1D-CNN) with 84.3% accuracy.

Conclusion

The use of PMs using machine learning techniques for detecting OSA events still has a long way to go. However, our method for developing explainable DL models demonstrates that PMs appear to be a promising alternative to PSG in the future for the detection of obstructive apnea events and the automatic calculation of AHI.

Validity study of a multiscaled fusion network using single-lead electrocardiogram signals for obstructive sleep apnea diagnosis

STUDY OBJECTIVES

We evaluated the validity of a squeeze-and-excitation and multiscaled fusion network (SE-MSCNN) using single-lead electrocardiogram (ECG) signals for obstructive sleep apnea detection and classification.

METHODS

Overnight polysomnographic data from 436 participants at the Sleep Center of the First Affiliated Hospital of Sun Yat-sen University were used to generate a new FAH-ECG dataset comprising 260, 88, and 88 single-lead ECG signal recordings for training, validation, and testing, respectively. The SE-MSCNN was employed for detection of apnea-hypopnea events from the acquired ECG segments. Sensitivity, specificity, accuracy, and F1 scores were assigned to assess algorithm performance. We also used the SE-MSCNN to estimate the apnea-hypopnea index, classify obstructive sleep apnea severity, and compare the agreement between 2 sleep technicians.

RESULTS

The SE-MSCNN's accuracy, sensitivity, specificity, and F1 score on the FAH-ECG dataset were 86.6%, 83.3%, 89.1%, and 0.843, respectively. Although slightly inferior to previously reported results using public datasets, it is superior to state-of-the-art open-source models. Furthermore, the SE-MSCNN had good agreement with manual scoring, such that the Spearman's correlations for the apnea-hypopnea index between the SE-MSCNN and 2 technicians were 0.93 and 0.94, respectively. Cohen's kappa scores in classifying the SE-MSCNN and the 2 sleep technicians were 0.72 and 0.78, respectively.

CONCLUSIONS

In this study, we validated the use of the SE-MSCNN in a clinical environment, and despite some limitations the network appeared to meet the performance standards for generalizability. Therefore, updating algorithms based on single-lead ECG signals can facilitate the development of novel wearable devices for efficient obstructive sleep apnea screening.

CITATION

Yue H, Li P, Li Y, et al. Validity study of a multiscaled fusion network using single-lead electrocardiogram signals for obstructive sleep apnea diagnosis. J Clin Sleep Med. 2023;19(6):1017-1025.

Design of an Intelligent Decision Support System Applied to the Diagnosis of Obstructive Sleep Apnea

Obstructive sleep apnea (OSA), characterized by recurrent episodes of partial or total obstruction of the upper airway during sleep, is currently one of the respiratory pathologies with the highest incidence worldwide. This situation has led to an increase in the demand for medical appointments and specific diagnostic studies, resulting in long waiting lists, with all the health consequences that this entails for the affected patients. In this context, this paper proposes the design and development of a novel intelligent decision support system applied to the diagnosis of OSA, aiming to identify patients suspected of suffering from the pathology. For this purpose, two sets of heterogeneous information are considered. The first one includes objective data related to the patient's health profile, with information usually available in electronic health records (anthropometric information, habits, diagnosed conditions and prescribed treatments). The second type includes subjective data related to the specific OSA symptomatology reported by the patient in a specific interview. For the processing of this information, a machine-learning classification algorithm and a set of fuzzy expert systems arranged in cascade are used, obtaining, as a result, two indicators related to the risk of suffering from the disease. Subsequently, by interpreting both risk indicators, it will be possible to determine the severity of the patients' condition and to generate alerts. For the initial tests, a software artifact was built using a dataset with 4400 patients from the Álvaro Cunqueiro Hospital (Vigo, Galicia, Spain). The preliminary results obtained are promising and demonstrate the potential usefulness of this type of tool in the diagnosis of OSA.

Sparse Decomposition of Heart Rate Using a Bernoulli-Gaussian Model: Application to Sleep Apnoea Detection

In this paper, we propose a sparse decomposition of the heart rate during sleep with an application to apnoea-RERA detection. We observed that the tachycardia following an apnoea event has a quasi-deterministic shape with a random amplitude. Accordingly, we model the apnoea-perturbed heart rate as a Bernoulli-Gaussian (BG) process convolved with a deterministic reference signal that allows the identification of tachycardia and bradycardia events. The problem of determining the BG series indicating the presence or absence of an event and estimating its amplitude is a deconvolution problem for which sparsity is imposed. This allows an almost syntactic representation of the heart rate on which simple detection algorithms are applied.

Design and Conceptual Proposal of an Intelligent Clinical Decision Support System for the Diagnosis of Suspicious Obstructive Sleep Apnea Patients from Health Profile

Obstructive Sleep Apnea (OSA) is a chronic sleep-related pathology characterized by recurrent episodes of total or partial obstruction of the upper airways during sleep. It entails a high impact on the health and quality of life of patients, affecting more than one thousand million people worldwide, which has resulted in an important public health concern in recent years. The usual diagnosis involves performing a sleep test, cardiorespiratory polygraphy, or polysomnography, which allows characterizing the pathology and assessing its severity. However, this procedure cannot be used on a massive scale in general screening studies of the population because of its execution and implementation costs; therefore, causing an increase in waiting lists which would negatively affect the health of the affected patients. Additionally, the symptoms shown by these patients are often unspecific, as well as appealing to the general population (excessive somnolence, snoring, etc.), causing many potential cases to be referred for a sleep study when in reality are not suffering from OSA. This paper proposes a novel intelligent clinical decision support system to be applied to the diagnosis of OSA that can be used in early outpatient stages, quickly, easily, and safely, when a suspicious OSA patient attends the consultation. Starting from information related to the patient's health profile (anthropometric data, habits, comorbidities, or medications taken), the system is capable of determining different alert levels of suffering from sleep apnea associated with different apnea-hypopnea index (AHI) levels to be studied. To that end, a series of automatic learning algorithms are deployed that, working concurrently, together with a corrective approach based on the use of an Adaptive Neuro-Based Fuzzy Inference System (ANFIS) and a specific heuristic algorithm, allow the calculation of a series of labels associated with the different levels of AHI previously indicated. For the initial software implementation, a data set with 4600 patients from the Álvaro Cunqueiro Hospital in Vigo was used. The results obtained after performing the proof tests determined ROC curves with AUC values in the range 0.8-0.9, and Matthews correlation coefficient values close to 0.6, with high success rates. This points to its potential use as a support tool for the diagnostic process, not only from the point of view of improving the quality of the services provided, but also from the best use of hospital resources and the consequent savings in terms of costs and time.

Multi-task learning neural networks for breath sound detection and classification in pervasive healthcare

With the emergence of many grave Chronic obstructive pulmonary diseases (COPDs) and the COVID-19 pandemic, there is a need for timely detection of abnormal respiratory sounds, such as deep and heavy breaths. Although numerous efficient pervasive healthcare systems have been proposed for tracking patients, few studies have focused on these breaths. This paper presents a method that supports physicians in monitoring in-hospital and at-home patients by monitoring their breath. The proposed method is based on three deep neural networks in audio analysis: RNNoise for noise suppression, SincNet - Convolutional Neural Network, and Residual Bidirectional Long Short-Term Memory for breath sound analysis at edge devices and centralized servers, respectively. We also developed a pervasive system with two configurations: (i) an edge architecture for in-hospital patients; and (ii) a central architecture for at-home ones. Furthermore, a dataset, named BreathSet, was collected from 27 COPD patients being treated at three hospitals in Vietnam to verify our proposed method. The experimental results demonstrated that our system efficiently detected and classified breath sounds with F1-scores of 90% and 91% for the tiny model version on low-cost edge devices, and 90% and 95% for the full model version on central servers, respectively. The proposed system was successfully implemented at hospitals to help physicians in monitoring respiratory patients in real time.

Breath monitoring, sleep disorder detection, and tracking using thin-film acoustic waves and open-source electronics

Apnoea, a major sleep disorder, affects many adults and causes several issues, such as fatigue, high blood pressure, liver conditions, increased risk of type II diabetes, and heart problems. Therefore, advanced monitoring and diagnosing tools of apnoea disorders are needed to facilitate better treatment, with advantages such as accuracy, comfort of use, cost effectiveness, and embedded computation capabilities to recognise, store, process, and transmit time series data. In this work we present an adaptation of our apnoea-Pi open-source surface acoustic wave (SAW) platform (Apnoea-Pi) to monitor and recognise apnoea in patients. The platform is based on a thin-film SAW device using bimorph ZnO and Al structures, including those fabricated as Al foils or plates, to achieve breath tracking based on humidity and temperature changes. We applied open-source electronics and provided embedded computing characteristics for signal processing, data recognition, storage, and transmission of breath signals. We show that the thin-film SAW device out-performed standard and off-the-shelf capacitive electronic sensors in terms of their response and accuracy for human breath-tracking purposes. This in combination with embedded electronics makes a suitable platform for human breath monitoring and sleep disorder recognition.

Detection of Sleep Apnea from Electrocardiogram and Pulse Oximetry Signals Using Random Forest

Sleep apnea (SA) is a common sleep disorder which could impair the human physiological system. Therefore, early diagnosis of SA is of great interest. The traditional method of diagnosing SA is an overnight polysomnography (PSG) evaluation. When PSG has limited availability, automatic SA screening with a fewer number of signals should be considered. The primary purpose of this study is to develop and evaluate a SA detection model based on electrocardiogram (ECG) and blood oxygen saturation (SpO2). We adopted a multimodal approach to fuse ECG and SpO2 signals at the feature level. Then, feature selection was conducted using the recursive feature elimination with cross-validation (RFECV) algorithm and random forest (RF) classifier used to discriminate between apnea and normal events. Experiments were conducted on the Apnea-ECG database. The introduced algorithm obtained an accuracy of 97.5%, a sensitivity of 95.9%, a specificity of 98.4% and an AUC of 0.992 in per-segment classification, and outperformed previous works. The results showed that ECG and SpO2 are complementary in detecting SA, and that the combination of ECG and SpO2 enhances the ability to diagnose SA. Therefore, the proposed method has the potential to be an alternative to conventional detection methods.

Apnea Detection in Polysomnographic Recordings Using Machine Learning Techniques

Sleep disorders are diagnosed in sleep laboratories by polysomnography, a multi-parameter examination that monitors biological signals during sleep. The subsequent evaluation of the obtained records is very time-consuming. The goal of this study was to create an automatic system for evaluation of the airflow and SpO2 channels of polysomnography records, through the use of machine learning techniques and a large database, for apnea and desaturation detection (which is unusual in other studies). To that end, a convolutional neural network (CNN) was designed using hyperparameter optimization. It was then trained and tested for apnea and desaturation. The proposed CNN was compared with the commonly used k-nearest neighbors (k-NN) method. The classifiers were designed based on nasal airflow and blood oxygen saturation signals. The final neural network accuracy for apnea detection reached 84%, and that for desaturation detection was 74%, while the k-NN classifier reached accuracies of 83% and 64% for apnea detection and desaturation detection, respectively.