Accuracy of deep learning in diagnosis of apnea syndrome: a systematic review and meta-analysis

Abstract

Objectives:

This systematic review and meta-analysis was carried out to elucidate the accuracy of image-based deep learning (DL) methods in the real-time detection of obstructive sleep apnea syndrome (OSAS).

Methods:

A systematic search was conducted for studies published since database establishment up to September 25, 2025, across databases including PubMed, Embase, Web of Science, and the Cochrane Library. The included studies were assessed for risk of bias by using the QUADAS-2 tool. During this meta-analysis, a bivariate mixed-effects model was employed and only synthesized the results from the meta-analysis of the validation sets. Meanwhile, subgroup analyses were conducted based on the generation methods of the validation sets.

Results:

A total of 39 original studies were ultimately included, all of which constructed DL images derived from electrocardiogram (ECG) images. Our meta-analysis results suggested that for the comprehensive validation set, the sensitivity, specificity, diagnostic odds ratio (DOR), and the area under summary receiver operating characteristic (SROC) curve were 0.93 (95% CI: 0.90–0.96), 0.95 (95% CI: 0.92–0.96), 252 (95% CI:116–549), and 0.98 (95% CI: 0.42–1.00), respectively. For the independent validation set, the sensitivity, specificity, and SROC curve were 0.93 (95% CI: 0.88–0.96), 0.95 (95% CI: 0.92–0.97), and 0.98 (95% CI: 0.42–1.00), respectively. For the K-fold cross-validation set, the sensitivity, specificity, positive likelihood ratio (LR), and SROC curve were 0.94 (95% CI: 0.88–0.97), 0.94 (95% CI: 0.89–0.96), 15.0 (95% CI: 8.1–27.6) and 0.98 (95% CI: 0.65–1.00), respectively.

Conclusion:

The ECGs-based DL models demonstrate ideal accuracy for the detection of OSAS and appear to be a viable method for real-time detection. During our research process, we found that the modeling was actually based on extracting studies from segments of ECGs, but the extracted segments appeared to vary in duration. Since this aspect was not subjected to subgroup analysis in our study, we plan to conduct further exploration and validation in subsequent research.

Systematic review registration:

CRD42023465176, https://www.crd.york.ac.uk/PROSPERO/home.

Introduction

OSAS is a clinical syndrome characterized by recurrent upper respiratory tract obstruction during sleep due to various causes, which leads to fragmented sleep and intermittent hypoxia during sleep periods (1). Studies have found that in patients with hypertension, coronary artery disease, heart failure, pulmonary hypertension, atrial fibrillation, and stroke, the prevalence of OSAS is as high as 40 to 80% (2). Its characteristics include repeated partial or complete respiratory pauses due to upper respiratory tract obstruction, affecting ventilation during sleep (3). Obstructive sleep apnea is very common among patients with cardiovascular diseases and is associated with the incidence and prevalence of hypertension, arrhythmias, coronary heart disease, heart failure, and stroke (4). It was found that the number of individuals affected by OSAS remains high, with an estimated 1 billion people suffering from OSAS worldwide (5).

As of now, the diagnostic process for OSAS is quite laborious. The gold standard method for diagnosing OSAS requires a polysomnography (PSG) system, a hospital setting, technical personnel, and a specialist physician (6). PSG involves recording, analyzing, and simultaneously collecting changes in multiple physiological signals, typically including but not limited to ECG and respiratory signals (6). However, a PSG device involves nearly 60 electrodes, making it difficult for a patient to have a normal sleep experience with so many connecting cables. Therefore, it becomes especially important to employ a method that not only offers high diagnostic reliability but is also simpler and more comfortable for patients (7).

In recent years, AI has begun to emerge and develop rapidly. In 2022, the highest investment was directed toward the healthcare and medical care sectors (8). Many anticipate that AI will achieve similar successes in the health sector, especially in diagnostics. Some believe that AI applications might even replace entire medical disciplines or create new roles to assist physicians (9). Against this background, medical image-based DL methods have gradually garnered extensive attention from researchers in clinical practice. For instance, in oncology, DL methods are employed to the identification of the benign or malignant nature of tumors based on pathological images, including early cancer detection, diagnosis, tumor classification and grading, molecular characterization, prognosis prediction, treatment response prediction, personalized treatment, automated radiation therapy workflows, discovery of novel anticancer drugs, and support in clinical trials (10, 11). Furthermore, we have noted other studies on DL and its application in diagnosing various types of diseases (12, 13).

In this context, some researchers have attempted to develop ECG-based DL models for detecting OSAS. However, the results from these DL approaches remain controversial, presenting certain challenges for the development of AI in this field. Therefore, this systematic review and meta-analysis was carried out to investigate the effectiveness and safety of DL methods based on medical images for real-time monitoring of OSAS, and to provide evidence-based opinions for the development and update of future real-time monitoring tools, such as wearable devices.

Methods

Study registration

The systematic review and meta-analysis adhered to the PRISMA, PRISMA NMA, and DMA guidelines, and prospectively registered on PROSPERO (ID: CRD42023465176).

Eligibility criteria

Inclusion criteria

• Studies such as case–control, cohort, case-cohort, nested case–control, and cross-sectional studies.

• Studies that have fully constructed DL models for diagnosing OSA.

• Some studies did not use other datasets for validation of the constructed models, and only performed cross-validation. These studies were also included in this systematic review.

• Different DL studies were based on the same dataset. These studies were also included in our systematic review.

• Included studies were original research reported in English.

Exclusion criteria

• Studies such as meta-analyses, literature reviews, conference papers, guidelines, and expert opinions.

• Given that this is a study on DL, which places a greater emphasis on the performance in the validation set, studies that did not perform any form of validation were excluded from our research.

• Studies lacking an assessment of DL model accuracy with the following outcome measures: ROC curve, C-index, sensitivity, specificity, accuracy, precision, confusion matrix, F1 score, and calibration curve.

• Studies focused on image segmentation.

• Studies with populations being neonates and children.

Data sources and search strategy

A systematical search was done for relevant literature published until September 25, 2025, in PubMed, Embase, Cochrane, and Web of Science databases, using MeSH + free-text terms, without limiting publication region or year (Table S1 for search strategy). To minimize the risk of missing newly published literature, we conducted a supplementary search of the databases on September 25, 2025.

Study selection and data extraction

Obtained articles were imported into EndNote. After excluding duplicates, original studies that initially met the criteria were reviewed by titles and abstracts. Full texts of preliminary eligible articles were downloaded for further screening to select the final original studies for this review.

Prior to extracting data, a standardized data extraction spreadsheet was developed, including title, the first author, year of publication, author’s country, study type, patient source, image source, number of patients with sleep apnea syndrome, the total number of cases involved in the generation of the validation set, number of cases of sleep apnea syndrome in the validation set, model type used, and whether a comparison with clinical physicians was made.

The literature selection and data extraction were independently done by two researchers (Alimila Saiyitijiang; Zhihui Nai) and cross-checked. In the event of any disputes, a third reviewer was engaged for resolution.

Risk of bias in studies

The risk of bias and applicability of the included studies were evaluated by adopting the QUADAS-2 tool (14). This tool evaluates four domains: patient selection, index test, reference standard, and flow and timing. Each domain contains specific questions answered as “Yes,” “No,” or “Uncertain,” corresponding to a judgment of bias risk as “Low,” “High,” or “Uncertain,” respectively. Studies are considered at low risk of bias if all signaling questions in each domain are answered with “Yes.” Any “No” among the answers to the signaling questions indicates potential bias, and the evaluator must then judge the risk of bias according to the guidelines provided. An “Uncertain” rating refers to situations with insufficient information for a definitive judgment.

Synthesis methods

Data analysis was performed utilizing Stata 15.0. A bivariate mixed-effects model was employed for the meta-analysis. Pooled estimates of sensitivity, specificity, Positive likelihood ratio (PLR), Negative likelihood ratio (NLR), Diagnostic odds ratio (DOR), and corresponding 95% CIs, were calculated using this model. The area under the SROC curve was also estimated. Deek’s funnel plot was adopted to assess publication bias. Additionally, a nomogram was employed to evaluate the clinical applicability of DL. Throughout the analysis, the prevalence of OSA in the studies included was used as the prior probability. Furthermore, we conducted subgroup analyses on the methods of validation set generation (independent validation and k-fold cross-validation). p < 0.05 was deemed statistically significant.

Results

Study selection

From the various databases, we retrieved 3697 publications. After removing 2,102 duplicates, we further excluded articles based on the following criteria: not related to the subject of this study or not involving DL (1,435 articles), not written in English (18 articles), systematic reviews and guidelines as well as registries and those using other research methods (99 articles). Additionally, 4 articles were excluded due to being unable to download or incomplete data. Ultimately, 39 publications (15–53) were included in the study (Figure 1).

Figure 1

Study characteristics

These articles were published primarily between 2018 and 2023, and only one (48) was published in 2011; 21 of these articles were from China (16, 17, 20–23, 25, 26, 29, 32–34, 36–39, 44–46, 50, 53), 5 from India (27, 29–31, 52), 3 from the United States (24, 40, 51), 3 from Iran (15, 19, 35), 2 from Korea (18, 41), 1 from Australia (42), 1 from Bangladesh (43), 1 from Jordan (48), and 1 from Turkey (49). Data from these articles were derived primarily from the 2000 Cardiology Association Physical Apnea-ECG Database, the PhysioNet publicly available Apnea-ECG Database, the UCDDB, and the Philips University Physical Apnea-ECG Database. In these studies, only 2 articles (33, 53) used 5-min segments, 1 (28) used 3-min segments, and 1 (26) used 6-min segments, while the rest of the articles used 1-min segments. Fourteen articles (15, 16, 22, 26, 28, 29, 31–33, 38, 44, 47–49) employed internal random sampling for validation, 13 (18, 19, 23, 24, 27, 30, 35, 36, 39, 40, 50–52) used K-fold cross-validation, and 13 (17, 19, 21, 25, 34, 37, 41–43, 45, 46, 51, 52) utilized external multicenter validation (Table 1).

Risk of bias in studies

Utilizing the QUADAS-2 tool, we primarily assessed the overall risk of bias and concerns regarding applicability. In terms of overall bias risk, regarding case selection, none of the studies avoided case–control designs. However, since these studies primarily utilized ECG-based DL, where the modeling variables do not involve manual coding, we believed this introduced only a minimal risk of bias. Moreover, we considered all exclusions of cases to be appropriate; thus, from the perspective of case selection, those studies were judged to have a low risk of bias. As for the index test, most studies did not describe or provide information on whether the interpretation of outcomes was done without knowledge of the reference standard results, and there was no specific threshold established. Since both factors are unlikely to affect the results of DL, we considered the risk of bias from the index test perspective as low. All included studies were able to accurately distinguish the target disease states. The use of blinding in interpreting the reference standard results was not mentioned, but considering its minimal impact on DL models, we viewed the implementation and interpretation of the reference standard as having a low risk of bias. Most studies did not explain the time interval between the index test and the reference standard, but given that OSAS is a chronic condition, we assessed this as having a low risk of bias. Since all patients underwent the same reference standard and all cases meeting the inclusion criteria were analyzed, we considered the risk of bias in terms of the flow and timing of patients as low. Since the relevant included patients and backgrounds were well matched with the evaluation questions, and the reference standard is highly applicable, we assessed the applicability concerns related to patient selection and the reference standard as low. For independent validation sets, the index test’s implementation and interpretation matched the evaluation questions well, suggesting a low risk. However, for K-fold cross-validation, the implementation and interpretation of the index test presented a high risk of bias in matching the evaluation questions (Figures 2, 3).

Figure 2

Figure 3

Meta-analysis

Synthesized results

The outcomes of our meta-analysis employing a bivariate mixed-effects model showed that the comprehensive validation set had a sensitivity of 0.93 (95% CI: 0.90–0.96), specificity of 0.95 (95% CI: 0.92–0.96), PLR of 17.7 (95% CI: 11.8–26.7), NLR of 0.07 (95% CI: 0.05–0.11), DOR of 252 (95% CI: 116–549), I² = 99.76 (99.74–99.77) and an area under the SROC curve of 0.98 (95% CI: 0.42–1.00) (Figures 4, 5).

Figure 4

Figure 5

In the studies included, about 30% of the ECG image segments were associated with OSAS. Using this data as a hypothesis for the prior probability of OSAS, when the DL judgment result was positive, the probability of the true result for sleep apnea syndrome was 88%; when the DL judgment result was negative, the probability of the true result not being sleep apnea syndrome was 97% (Figure 6). Deek’s funnel plot indicated no significant publication bias across the studies (Figure 7).

Figure 6

Figure 7

Subgroup analysis

The sensitivity, specificity, PLR, NLR, DOR and SROC curve of the independent validation set were 0.93 (95% CI: 0.88–0.96), 0.95 (95% CI: 0.92–0.97), 19.5 (95% CI: 11.3–33.7), 0.07 (95% CI: 0.04–0.12), 274 (95% CI: 101–743), and 0.98 (95% CI: 0.42–1.00), respectively (Supplementary Figures S1, S2) and I² = 99.76 (95% CI: 99.74–99.77). When the DL results were positive, the probability that the true result for apnea syndrome was 88%. When the DL judgment result was negative, the probability of the true result not being sleep apnea syndrome was 97% (Supplementary Figure S3). Deek’s funnel plot indicated no significant publication bias across the studies (Supplementary Figure S4).

The sensitivity, specificity, PLR, NLR, DOR and SROC curve of the K-fold cross-validation set were 0.94 (95% CI: 0.88–0.97), 0.94 (95% CI: 0.89–0.96), 15.0 (95% CI: 8.1–27.6), 0.07 (95% CI: 0.03–0.13), 227 (95% CI: 64–808), and 0.98 (95% CI: 0.65–1.00), I² = 99.76 (99.74–99.77) respectively (Supplementary Figures S5, S6). When the DL results were positive, the probability that the true result for apnea syndrome was 90%. When the DL judgment result was negative, the probability of the true result not being sleep apnea syndrome was 96% (Supplementary Figure S7). Deek’s funnel plot revealed no significant publication bias across the studies (Supplementary Figure S8).

Discussion

In this systematic review and meta-analysis, 39 original studies were analyzed using a bivariate mixed-effects model, synthesizing solely the results of the meta-analysis of the validation sets. From the analysis, it was observed that the sensitivity and specificity for real-time OSAS detection using ECG image-based DL were 0.93 (95% CI: 0.90–0.96) and 0.95 (95% CI: 0.92–0.96), respectively. For the independent validation sets, the sensitivity and specificity of detecting OSAS with ECG image-based DL were 0.93 (95% CI: 0.88–0.96) and 0.95 (95% CI: 0.92–0.97), respectively. For the K-fold cross-validation sets, the sensitivity and specificity of detecting OSAS using ECG image-based DL were 0.94 (95% CI: 0.88–0.97) and 0.94 (95% CI: 0.89–0.96), respectively.

In this study, we also noticed that some researchers have concentrated on the detection of OSAS using other methods such as polysomnography (PSG), biomarkers, imaging, etc. For instance, the method of “balloon angiography” used by Huysmans et al. (54) involved installing motion sensors under the bed of sleeping individuals to record coarse body movements, respiratory-related movements, and even cardiac motion during PSG. The combination of these three signals can largely provide a relatively better assessment of sleep state margins and sleep-related breathing disorders. However, this detection method is more complex and requires a higher level of environmental monitoring. However, the study results showed a screening sensitivity of 0.77 and specificity of 0.62 for patients with severe apneas, and for general apnea patients, a screening sensitivity of 0.72 and specificity of 0.70, both lower than those found in our study.

A study conducted by Zorlu D et al. (55) investigated the use of complete blood count parameters to predict the OSAS diagnosis and to grade the severity of OSAS. The study found that a lymphocyte value of 737.14 as a cutoff point showed a sensitivity (90.7%) and specificity (92.6%), indicating that the lymphocyte variable possesses a certain diagnostic value for mild OSAS. No significant cutoff values could be identified for moderate and severe OSAS groups, as the area under the ROC curve was not significant (p > 0.05) for these groups. Blood cell data are influenced by various factors, although eliminating the impact of patient-related factors on NLR, PLR, and WMR parameters was a notable strength in their study, they did not evaluate the impact of comorbidities, a history of corticosteroid treatments, or other inflammatory factors like calcitonin on complete blood count parameters.

Mahesh N et al. (56) also conducted a meta-analysis on diagnosing OSAS using the STOP-Bang questionnaire validated with polysomnography. In the sleep clinic population, the sensitivities for detecting any OSAS (AHI > 5), moderate to severe OSAS (AHI > 15), as well as severe OSAS (AHI > 30) were 90, 94, and 96%, respectively. Similarly, this pattern was noted in the surgical patient population. As the STOP-Bang score escalates, the likelihood of moderate and severe OSAS occurrence rises correspondingly. However, this study demonstrated moderate to high heterogeneity in the systematic review and meta-analysis, one factor of which was the variability of target populations in various studies and potential differences in the prevalence of OSAS among diverse populations. Additionally, there is a dearth of confirmatory studies in surgical patients. Moreover, it failed to achieve real-time detection, and the detection process was time-consuming.

In our study, we found that ECG-based methods for diagnosing OSA demonstrated highly favorable accuracy. ECG-based DL methods also achieved excellent diagnostic accuracy for OSA. However, significant challenges remain in the detection of OSA. For instance, OSAS is a sleep disorder characterized by repetitive cessation of airflow lasting at least 10 s per event, with each apnea episode accompanied by cardiovascular changes. During apnea, heart rate often decreases, and when breathing resumes, a relative tachycardia can be observed. Blood pressure drops during apnea episodes and rises at the end of apnea as sympathetic nerve activity increases. Oxygen saturation declines with the cessation of breathing, reaching its lowest point after several cycles of resumed breathing. The characteristics of fluctuations in blood oxygen saturation and heart rate are utilized by portable diagnostic devices for the early detection of sleep apnea (57).

Hypoxemia and hypercapnia episodes during sleep apnea can trigger physiological responses, including activation of the sympathetic nervous system, oxidative stress, inflammation, and endothelial dysfunction. Excessive daytime sleepiness may be described by patients as fatigue, low energy, or difficulty concentrating, reflecting an inability to maintain full wakefulness or alertness during the wakeful portion of the sleep–wake cycle (58). Studies have also indicated a certain association between OSA and traffic accidents. However, the relationship between OSA and traffic incidents is often complex and multifactorial in etiology, necessitating further investigation into the potential causes of these events. The process of screening for OSA solely through ECG measurements remains limited (59).

In our study of the ECG image-based DL models, the validation approach presented the true detection performance of AI for OSAS. In the included studies, 26 used an independent validation set to provide results, and 13 used K-fold cross-validation to provide results. According to the results, the sensitivity and specificity for real-time OSAS detection by ECG image-based DL were 0.93 (95% CI: 0.90–0.96) and 0.95 (95% CI: 0.92–0.96), respectively. For the independent validation sets, the sensitivity and specificity of detecting OSAS with ECG image-based DL were 0.93 (95% CI: 0.88–0.96) and 0.95 (95% CI: 0.92–0.97), respectively. For the K-fold cross-validation sets, the sensitivity and specificity of detecting OSAS using ECG image-based DL were, respectively, 0.94 (95% CI: 0.88–0.97) and 0.94 (95% CI: 0.89–0.96). The performance of independent validation sets and K-fold cross-validation showed no significant differences, indicating that DL methods for OSAS detection are highly stable. Moreover, among the included studies, Feiteng (22) designed a user-friendly OSAS monitoring system equipped with multimedia devices for accurate and efficient OSAS detection in intelligent healthcare management. In the study by Hemrajani, P (31), a compact, accurate, and portable wearable device, Sleepify, was developed to address the cumbersome and time-consuming nature of PSG, allowing patients to comfortably wear the device at home, record their ECG signals, and detect sleep apnea events, with the device alerting them to any incidents during the night.

Advantages and limitations of the study

The DL models we investigated analyze long-duration single-lead ECG records of patients, and use one-dimensional ECG signals as input to detect apnea events, which also involve training and validation of the collected data. This approach does not involve QRS complex detection or analysis of RR intervals or cardiac function and requires less labor compared to traditional diagnostic methods, offering greater convenience and high accuracy.

In comparison with non-traditional detection methods, such as polysomnography (PSG) and blood biomarkers, our method is more straightforward in operation. Nevertheless, it is undeniable that our detection method is influenced by numerous factors. These factors encompass the instruments employed and the proficiency of the testing personnel. Additionally, given that the data utilized is sourced from public databases, there are indeed certain constraints in this respect.

As this is a retrospective study, it is seriously lacking in the explanation of blinding. In this deep learning, there are inherent rules for determining the threshold. We believe that it used the threshold rather than set the threshold probability. Therefore, we think that this does not bring an excessively high risk of bias to the entire training process.

Nevertheless, the specificity in our study population remains a challenge, mainly due to our modeling data primarily coming from four public databases: the 2000 Cardiology Association Physical Apnea-ECG Database, the PhysioNet publicly available Apnea-ECG Database, the UCDDB, and the Philips University Physical Apnea-ECG Database. These databases used different units for ECG recording, such as 1 min, 3 min (28), 5 min (33, 53), and 6 min ECG recording segments (26), the majority of them developed models based on segments with a duration of 1 min. Only a small number of models were constructed using segments of 5 min, 3 min, and 6 min. Moreover, some of these studies utilized cross - validation. As a result, it was challenging for us to perform further subgroup analysis. This also represents a limitation of our present study.

In addition, since the diagnosis of OSA was not clearly marked in most of the studies we included, we were unable to further present precise information about the diagnostic process in the original literature.

Additionally, the data included in our manuscript is not comprehensive; for instance, data from regions such as South America and New England are not represented. While there is a noted association between OSA and traffic accidents, the etiological relationship is often complex and multifactorial, requiring further investigation into the potential causes of these incidents. The screening process based solely on ECG measurements remains insufficient. Furthermore, some studies have highlighted the role of nocturnal blood oxygen saturation in the diagnosis of OSA. However, due to the limitations of the included literature, our study did not incorporate nocturnal blood pressure or nocturnal blood oxygen saturation measurements.

Also, We endeavored to utilize external validation to explore the influence of the same dataset across different studies on our research outcomes. Nevertheless, the scarcity of external validation datasets has impeded our ability to conduct a more in - depth analysis in this regard.

Among the 39 studies in our review, 33 used a 1-min window, one used a 3-min window, two used a 5-min window, one used a 6-min window, and two did not explicitly report the segment duration. For the meta-analysis, we employed a bivariate random-effects model, which requires at least four studies providing 2 × 2 contingency tables for diagnostic accuracy. Due to the limited number of studies in each subgroup defined by segment duration, we were unable to perform stratified subgroup analyses. This represents a limitation of our study.

Due to the existence of diverse frameworks in deep learning and the use of the bivariate mixed-effects model, at least four studies are required. Among the included studies, there are significant differences in these deep learning frameworks, and some have even conducted partial ablation experiments. Therefore, we were unable to further discuss the detection performance of different deep learning methods in this aspect. We directly summarized them as ideal diagnostic tools. This is also a major approach for discussing the detection of diseases by deep learning based on a single systematic review.

Conclusion

The results from this study suggest that medical image-based DL methods have demonstrated marked efficacy and safety profiles in real-time monitoring of OSAS, and the use of AI for the diagnosis of OSAS is a novel and effective diagnostic approach, which can provide reliable evidence for future development and design of real-time monitoring tools for OSAS.

References

• 1.

  Drager LF Togeiro SM Polotsky VY Lorenzi-Filho G . Obstructive sleep apnea: a cardiometabolic risk in obesity and the metabolic syndrome. J Am Coll Cardiol. (2013) 62:569–76. doi: 10.1016/j.jacc.2013.05.045

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 2.

  Yeghiazarians Y Jneid H Tietjens JR Redline S Brown DL El-Sherif N et al . Obstructive sleep Apnea and cardiovascular disease: a scientific statement from the American Heart Association. Circulation. (2021) 144:e56–67. doi: 10.1161/CIR.0000000000000988

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3.

  Mannarino MR Di Filippo F Pirro M . Obstructive sleep apnea syndrome. Eur J Intern Med. (2012) 23:586–93. doi: 10.1016/j.ejim.2012.05.013

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4.

  Javaheri S Barbe F Campos-Rodriguez F Dempsey JA Khayat R Javaheri S et al . Sleep Apnea: types, mechanisms, and clinical cardiovascular consequences. J Am Coll Cardiol. (2017) 69:841–58. doi: 10.1016/j.jacc.2016.11.069

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 5.

  Lyons MM Bhatt NY Pack AI Magalang UJ . Global burden of sleep-disordered breathing and its implications. Respirology. (2020) 25:690–702. doi: 10.1111/resp.13838

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6.

  Patil SP . What every clinician should know about polysomnography. Respir Care. (2010) 55:1179–95.

  • Google Scholar

• 7.

  Bilgin C Erkorkmaz U Ucar MK Akin N Nalbant A Annakkaya AN . Use of a portable monitoring device (Somnocheck Micro) for the investigation and diagnosis of obstructive sleep apnoea in comparison with polysomnography. Pak J Med Sci. (2016) 32:471–5. doi: 10.12669/pjms.322.9561

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 8.

  Maslej N Fattorini L Brynjolfsson E Etchemendy J Ligett K Lyons T et al . The AI index 2023 annual report. Stanford, CA: Stanford University (2023).

  • Google Scholar

• 9.

  Coiera E . The fate of medicine in the time of AI. Lancet. (2018) 392:2331–2. doi: 10.1016/S0140-6736(18)31925-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 10.

  Chen ZH Lin L Wu CF Li CF Xu RH Sun Y . Artificial intelligence for assisting cancer diagnosis and treatment in the era of precision medicine. Cancer Commun. (2021) 41:1100–15. doi: 10.1002/cac2.12215

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11.

  Bhinder B Gilvary C Madhukar NS Elemento O . Artificial intelligence in Cancer research and precision medicine. Cancer Discov. (2021) 11:900–15. doi: 10.1158/2159-8290.CD-21-0090

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 12.

  Shen D Wu G Suk HI . Deep learning in medical image analysis. Annu Rev Biomed Eng. (2017) 19:221–48. doi: 10.1146/annurev-bioeng-071516-044442

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13.

  Qiu S Miller MI Joshi PS Lee JC Xue C Ni Y et al . Multimodal deep learning for Alzheimer's disease dementia assessment. Nat Commun. (2022) 13:3404. doi: 10.1038/s41467-022-31037-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14.

  Whiting PF Rutjes AW Westwood ME Mallett S Deeks JJ Reitsma JB et al . QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. (2011) 155:529–36. doi: 10.7326/0003-4819-155-8-201110180-00009

  • CrossRef
  • Google Scholar

• 15.

  Bahrami M Forouzanfar M . Detection of sleep Apnea from single-Lead ECG: Comparison of deep learning algorithms. 2021 IEEE international symposium on medical measurements and applications (MeMeA); 2021IEEE: Switzerland

  • Google Scholar

• 16.

  Chang HY Yeh CY Lee CT Lin CC . A sleep Apnea detection system based on a one-dimensional deep convolution neural network model using single-Lead electrocardiogram. Sensors. (2020) 20:4157. doi: 10.3390/s20154157

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 17.

  Chen J Shen M Ma W Zheng W . A spatio-temporal learning-based model for sleep apnea detection using single-lead ECG signals. Front Neurosci. (2022) 16:972581. doi: 10.3389/fnins.2022.972581

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18.

  Ullah N Mahmood T Kim SG Nam SH Sultan H Park KR . DCDA-net: dual-convolutional dual-attention network for obstructive sleep apnea diagnosis from single-lead electrocardiograms. Eng Appl Artif Intell. (2023) 123:106451. doi: 10.1016/j.engappai.2023.106451

  • CrossRef
  • Google Scholar

• 19.

  Zarei A Beheshti H Asl BM . Detection of sleep apnea using deep neural networks and single-lead ECG signals. Biomed Signal Process Control. (2022) 71:103125. doi: 10.1016/j.bspc.2021.103125

  • CrossRef
  • Google Scholar

• 20.

  Yang Q Zou L Wei K Liu G . Obstructive sleep apnea detection from single-lead electrocardiogram signals using one-dimensional squeeze-and-excitation residual group network. Comput Biol Med. (2022) 140:105124. doi: 10.1016/j.compbiomed.2021.105124

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21.

  Wang Z Peng C Li B Penzel T Liu R Zhang Y et al . Single-lead ECG based multiscale neural network for obstructive sleep apnea detection. Inter Things. (2022) 20:100613. doi: 10.1016/j.iot.2022.100613

  • CrossRef
  • Google Scholar

• 22.

  Teng F Wang D Yuan Y Zhang H Singh AK Lv Z . Multimedia monitoring system of obstructive sleep apnea via a deep active learning model. IEEE Multi. (2022) 29:48–56. doi: 10.1109/MMUL.2022.3146141

  • CrossRef
  • Google Scholar

• 23.

  Setiawan F Lin CW . A deep learning framework for automatic sleep Apnea classification based on empirical mode decomposition derived from single-Lead electrocardiogram. Life. (2022) 12:1509. doi: 10.3390/life12101509

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 24.

  Paul T Hassan O Alaboud K Islam H Rana MKZ Islam SK et al . ECG and SpO(2) signal-based real-time sleep apnea detection using feed-forward artificial neural network. AMIA Jt Summits Transl Sci Proc. (2022) 2022:379–85.

  • Google Scholar

• 25.

  Qin H Liu G . A dual-model deep learning method for sleep apnea detection based on representation learning and temporal dependence. Neurocomputing. (2022) 473:24–36. doi: 10.1016/j.neucom.2021.12.001

  • CrossRef
  • Google Scholar

• 26.

  Hu S Cai W Gao T Wang M . A hybrid transformer model for obstructive sleep Apnea detection based on self-attention mechanism using single-Lead ECG. IEEE Trans Instrum Meas. (2022) 71:1–11. doi: 10.1109/TIM.2022.3193169

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 27.

  Gupta K Bajaj V Ansari IA . Osacn-net: automated classification of sleep apnea using deep learning model and smoothed Gabor spectrograms of ECG signal. IEEE Trans Instrum Meas. (2022) 71:1–9. doi: 10.1109/TIM.2021.3132072

  • CrossRef
  • Google Scholar

• 28.

  Liu H Cui S Zhao X Cong F . Detection of obstructive sleep apnea from single-channel ECG signals using a CNN-transformer architecture. Biomed Signal Process Control. (2023) 82:104581. doi: 10.1016/j.bspc.2023.104581

  • CrossRef
  • Google Scholar

• 29.

  Kumar Tyagi P Agrawal D . Automatic detection of sleep apnea from single-lead ECG signal using enhanced-deep belief network model. Biomed Signal Process Control. (2023) 80:104401. doi: 10.1016/j.bspc.2022.104401

  • CrossRef
  • Google Scholar

• 30.

  Kumar CB Mondal AK Bhatia M Panigrahi BK Gandhi TK . Self-supervised representation learning-based OSA detection method using Single-Channel ECG signals. IEEE Trans Instrum Meas. (2023) 72:1–15. doi: 10.1109/TIM.2023.3261931

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31.

  Hemrajani P Dhaka VS Rani G Shukla P Bavirisetti DP . Efficient deep learning based hybrid model to detect obstructive sleep apnea. Sensors. (2023) 23:104692. doi: 10.3390/s23104692

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32.

  Chen X Ma W Gao W Fan X . BAFNet: bottleneck attention based fusion network for sleep apnea detection. IEEE J Biomed Health Inform. (2023) 28:1–12. doi: 10.1109/JBHI.2023.3278657

  • CrossRef
  • Google Scholar

• 33.

  Chen X Chen Y Ma W Fan X Li Y . Toward sleep apnea detection with lightweight multi-scaled fusion network. Knowl-Based Syst. (2022) 247:108783. doi: 10.1016/j.knosys.2022.108783

  • CrossRef
  • Google Scholar

• 34.

  Cao K Lv X . Multi-task feature fusion network for obstructive sleep apnea detection using single-lead ECG signal. Measurement. (2022) 202:111787. doi: 10.1016/j.measurement.2022.111787

  • CrossRef
  • Google Scholar

• 35.

  Bahrami M Forouzanfar M . Deep learning forecasts the occurrence of sleep Apnea from single-Lead ECG. Cardiovasc Eng Technol. (2022) 13:809–15. doi: 10.1007/s13239-022-00615-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 36.

  Yu Y Yang Z You Y Shan W . FASSNet: fast apnea syndrome screening neural network based on single-lead electrocardiogram for wearable devices. Physiol Meas. (2021) 42:085005. doi: 10.1088/1361-6579/ac184e

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 37.

  Li K Pan W Li Y Jiang Q Liu G . A method to detect sleep apnea based on deep neural network and hidden Markov model using single-lead ECG signal. Neurocomputing. (2018) 294:94–101. doi: 10.1016/j.neucom.2018.03.011

  • CrossRef
  • Google Scholar

• 38.

  Wang L Lin Y Wang J . A RR interval based automated apnea detection approach using residual network. Comput Methods Prog Biomed. (2019) 176:93–104. doi: 10.1016/j.cmpb.2019.05.002

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39.

  Feng K Qin H Wu S Pan W Liu G . A sleep Apnea detection method based on unsupervised feature learning and single-Lead electrocardiogram. IEEE Trans Instrum Meas. (2021) 70:1–12. doi: 10.1109/TIM.2020.3017246

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40.

  Faust O Barika R Shenfield A Ciaccio EJ Acharya UR . Accurate detection of sleep apnea with long short-term memory network based on RR interval signals. Knowl-Based Syst. (2021) 212:106591. doi: 10.1016/j.knosys.2020.106591

  • CrossRef
  • Google Scholar

• 41.

  Urtnasan E Park JU Joo EY Lee KJ . Automated detection of obstructive sleep Apnea events from a single-Lead electrocardiogram using a convolutional neural network. J Med Syst. (2018) 42:104. doi: 10.1007/s10916-018-0963-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 42.

  Sharan RV Berkovsky S Xiong H Coiera E . ECG-derived heart rate variability interpolation and 1-D convolutional neural networks for detecting sleep Apnea. Annu Int Conf IEEE Eng Med Biol Soc. (2020) 2020:637–40. doi: 10.1109/EMBC44109.2020.9175998

  • CrossRef
  • Google Scholar

• 43.

  Mashrur FR Islam MS Saha DK Islam SMR Moni MA . SCNN: scalogram-based convolutional neural network to detect obstructive sleep apnea using single-lead electrocardiogram signals. Comput Biol Med. (2021) 134:104532. doi: 10.1016/j.compbiomed.2021.104532

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44.

  Zhang J Tang Z Gao J Lin L Liu Z Wu H et al . Automatic detection of obstructive sleep Apnea events using a deep CNN-LSTM model. Comput Intell Neurosci. (2021) 2021:5594733. doi: 10.1155/2021/5594733

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45.

  Fang H Lu C Hong F Jiang W Wang T . Sleep apnea detection based on multi-scale residual network. Life. (2022) 12:119. doi: 10.3390/life12010119

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 46.

  Shen Q Qin H Wei K Liu G . Multiscale deep neural network for obstructive sleep Apnea detection using RR interval from single-Lead ECG signal. IEEE Trans Instrum Meas. (2021) 70:1–13. doi: 10.1109/TIM.2021.3062414

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47.

  Thompson S Fergus P Chalmers C Reilly D Detection of obstructive sleep Apnoea using features extracted from segmented time-series ECG signals using a one dimensional convolutional neural network. 2020IEEE Access (99):1 doi:10.1109/ACCESS.2023.3346689

  • CrossRef
  • Google Scholar

• 48.

  Lweesy K Fraiwan L Khasawneh N Dickhaus H . New automated detection method of OSA based on artificial neural networks using P-wave shape and time changes. J Med Syst. (2011) 35:723–34. doi: 10.1007/s10916-009-9409-z

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 49.

  Nasifoglu H Erogul O . Obstructive sleep apnea prediction from electrocardiogram scalograms and spectrograms using convolutional neural networks. Physiol Meas. (2021) 42:065010. doi: 10.1088/1361-6579/ac0a9c

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 50.

  Niroshana SMI Zhu X Nakamura K Chen W . A fused-image-based approach to detect obstructive sleep apnea using a single-lead ECG and a 2D convolutional neural network. PLoS One. (2021) 16:e0250618. doi: 10.1371/journal.pone.0250618

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51.

  Sheta AF Turabieh H Thaher T Too J Mafarja MM Hossain MS et al . Diagnosis of obstructive sleep apnea from ECG signals using machine learning and deep learning classifiers. Appl Sci. (2021) 11:6622. doi: 10.3390/app11146622

  • CrossRef
  • Google Scholar

• 52.

  Singh H Tripathy RK Pachori RB . Detection of sleep apnea from heart beat interval and ECG derived respiration signals using sliding mode singular spectrum analysis. Digit Signal Process. (2020) 104:102796. doi: 10.1016/j.dsp.2020.102796

  • CrossRef
  • Google Scholar

• 53.

  Wang T Lu C Shen G Hong F . Sleep apnea detection from a single-lead ECG signal with automatic feature-extraction through a modified LeNet-5 convolutional neural network. PeerJ. (2019) 7:e7731. doi: 10.7717/peerj.7731

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 54.

  Huysmans D Borzée P Testelmans D Buyse B Willemen T Huffel SV et al . Evaluation of a commercial ballistocardiography sensor for sleep apnea screening and sleep monitoring. Sensors. (2019) 19:133. doi: 10.3390/s19092133

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 55.

  Zorlu D Ozyurt S Bırcan HA Erturk A . Do complete blood count parameters predict diagnosis and disease severity in obstructive sleep apnea syndrome?Eur Rev Med Pharmacol Sci. (2021) 25:4027–36. doi: 10.26355/eurrev_202106_26044

  • CrossRef
  • Google Scholar

• 56.

  Nagappa M Liao P Wong J Auckley D Ramachandran SK Memtsoudis S et al . Validation of the STOP-bang questionnaire as a screening tool for obstructive sleep Apnea among different populations: a systematic review and Meta-analysis. PLoS One. (2015) 10:e0143697. doi: 10.1371/journal.pone.0143697

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 57.

  Roos M Althaus W Rhiel C Penzel T Peter JH von Wichert P . Comparative use of MESAM IV and polysomnography in sleep-related respiratory disorders. Pneumologie. (1993) 47:112–8.

  • Google Scholar

• 58.

  Slater G Steier J . Excessive daytime sleepiness in sleep disorders. J Thorac Dis. (2012) 4:608–16. doi: 10.3978/j.issn.2072-1439.2012.10.07

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 59.

  Felix M Intriago Alvarez MB Vanegas E Farfán Bajaña MJ Sarfraz Z Sarfraz A et al . Risk of obstructive sleep apnea and traffic accidents among male bus drivers in Ecuador: is there a significant relationship?Ann Med Surg. (2022) 74:103296. doi: 10.1016/j.amsu.2022.103296

  • Pubmed Abstract
  • CrossRef
  • Google Scholar