{"paper_id":"233f912e-b74a-4f2e-9e40-970329a9f9ee","body_text":"Sleep Apnea Detection Using Wearable ECG and Deep Learning: Validation with Polysomnography | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Sleep Apnea Detection Using Wearable ECG and Deep Learning: Validation with Polysomnography Sue Hyun Lee, Minwoo Kim, Sung Pil Cho, Yeewoong Kim, Min Kyung Chu, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7537553/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 15 You are reading this latest preprint version Abstract Obstructive sleep apnea (OSA) is a highly prevalent disorder that remains underdiagnosed due to the limited accessibility and high cost of polysomnography (PSG), the current diagnostic standard. This study presents a deep learning model that detects OSA from single-lead electrocardiogram (ECG) signals acquired via a wearable patch Holter device. We collected ECG data from 92 adult patients undergoing overnight PSG at a sleep clinic. A 1-dimensional dilated convolutional neural network (1D-CNN) was trained using 3-minute ECG segments and validated against PSG-derived apnea-hypopnea index (AHI) scores. The model achieved an accuracy of 81.1%, precision of 81.9%, recall of 82.5%, F1-score of 82.2%, and an area under the receiver operating characteristic curve (AUROC) of 0.875. The predicted AHI was strongly correlated with PSG AHI (r = 0.847), and the model outperformed conventional screening questionnaires such as STOP-Bang in identifying moderate-to-severe OSA (AUROC = 0.888). These results demonstrate the feasibility of using wearable ECG and deep learning for accurate and scalable OSA detection. This approach may offer a non-invasive and cost-effective alternative to PSG in both clinical and community-based screening settings. Health sciences/Cardiology Biological sciences/Computational biology and bioinformatics Health sciences/Diseases Health sciences/Health care Health sciences/Medical research Obstructive sleep apnea Electrocardiogram Deep learning Wearable devices Convolutional neural network Polysomnography validation Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Obstructive sleep apnea (OSA) is a common sleep-related breathing disorder characterized by repeated upper airway obstruction during sleep, resulting in intermittent hypoxia and sleep fragmentation. It has been associated with a wide range of adverse health outcomes, including hypertension, cardiovascular and cerebrovascular diseases, metabolic dysfunction, and impaired daytime functioning 1 – 5 . Epidemiological studies estimate that approximately 936 million adults worldwide are affected by mild-to-severe OSA, and over 400 million individuals suffer from moderate-to-severe forms, underscoring its significant public health impact 6 . Despite its prevalence, OSA remains underdiagnosed due to limitations in current diagnostic methods. Polysomnography (PSG), the clinical gold standard, is resource-intensive, time-consuming, and typically restricted to specialized sleep laboratories. Alternative approaches such as home sleep apnea testing and pulse oximetry offer greater accessibility but are often limited by reduced sensitivity and challenges in signal quality during natural sleep movement 7 – 10 . Recent advances in artificial intelligence (AI) and wearable biosensors have introduced new possibilities for OSA detection using physiological signals such as electrocardiograms (ECG) 11 . Specifically, deep learning models have shown promising performance in extracting apnea-relevant features directly from ECG signals, without the need for manual feature engineering. Wearable patch-type ECG devices further enhance the feasibility of out-of-hospital sleep monitoring due to their simplicity, comfort, and continuous data acquisition 12 , 13 . In this study, we developed and validated a deep learning model for sleep apnea detection using ECG signals collected from a wearable patch Holter device. The model was trained to classify apnea events based on 3-minute ECG segments and evaluated against PSG-derived apnea-hypopnea index (AHI) scores. Our findings suggest that this approach may offer a scalable, non-invasive solution for OSA screening, with potential applications in both clinical and community settings. Results Clinical characteristics The demographic and clinical characteristics of the study participants are shown in Table 1 . We recruited a cohort of 92 patients with an average age of 55.1 years. Men constituted a significant majority of the patients (75.0%, 69 patients). Our study population exhibited 26.0 ± 4.4 kg/m 2 for BMI, indicating an overweight status. In terms of OSA severity, the distribution was as follows: no OSA was observed in 12.0% of the patients (11 patients), while mild OSA was diagnosed in 12.0% (11 patients), moderate OSA in 28.3% (26 patients), and severe OSA in 47.8% (44 patients) of patients. Table 1 Demographic and Clinical Characteristics of Patients in the Study Characteristics Subjects, n = 92 Age (year) 55.1 ± 14.7 Gender (male), n (%) 69 (75%) Body mass index (kg/m 2 ) 26.0 ± 4.4 PSQI 8.0 (5–11) ISI 9.0 (4–14) ESS 6.0 (4–10) STOP-BANG 4.0 (3–5) BDI-II 10 (6–18) Time in bed (hours) 7.4 ± 0.8 Total sleep time (hours) 5.7 ± 1.2 Sleep efficiency (%) 76.9 ± 17.3 Arousal index (events/hour) 37.5 ± 18.8 O2 saturation nadir (%) 83.9 ± 7.5 Apnea-hypopnea index (events/hour) 32.9 ± 23.4 OSA severity No OSA, n (%) 11 (12.0%) Mild OSA, n (%) 11 (12.0%) Moderate OSA, n (%) 26 (28.3%) Severe OSA, n (%) 44 (47.8%) Data are presented as mean ± standard deviation for normally distributed quantitative data. For non-normally distributed quantitative data, data are presented as the medians with the interquartile range (25th–75th percentiles). BDI-II: Beck Depression Inventory-II; ESS: Epworth Sleepiness Scale; ISI: Insomnia Severity Index; OSA: Obstructive Sleep Apnea; PSQI: Pittsburgh Sleep Quality Index. Median values and interquartile ranges were calculated for the questionnaires. The median PSQI, ISI, ESS, and STOP-Bang scores were 8.0, 9.0, 6.0, and 4.0, respectively. These values indicate that our study population tended to have poor sleep quality, mild insomnia, normal levels of daytime sleepiness, and an intermediate risk of OSA. The median BDI-II score was 10.0, indicating that the patients were less likely to have depression. The mean value of time in bed was 7.4 ± 0.8 hours, and the total sleep time was 5.7 ± 1.2 hours. The sleep efficiency was 76.9 ± 17.3%, showing decreased sleep efficiency. However, this study was conducted in a hospital for overnight laboratory PSG recordings, which may have affected sleep efficiency. The arousal index was 37.5 ± 18.8 events/hour, which is a markedly increased value, even considering the first-night effect. The mean AHI was 32.9 ± 23.4 events/ hour, and the mean O2 saturation nadir was 83.9 ± 7.5%. Model performance in apnea detection We experimented with segments of different lengths based on the model configuration described earlier. We experimented with segments of 5 minutes, 3 minutes, 1 minute, and 30 seconds, and found that the 3-minute segment performed the best, as shown in the supplementary information table S2. The results of our experiments using the 3-minute segment are as follows. The total number of segments was 13,916, wherein 9,741 segments were used as the training set and 4,175 segments were used as the test set. The 9,741 training sets consisted of 4,649 normal segments and 5,092 apnea segments, whereas the 4,175 test sets consisted of 1,965 normal segments and 2,210 apnea segments. The final model showed an accuracy of 0.811, precision of 0.819, recall of 0.825, F1 score of 0.822, and AUROC of 0.875 (95% CI, 0.864–0.886) when evaluating the apnea detection performance on the test set (Fig. 1 ). Correlation and Bland-Altman Plots We observed a strong linear correlation (Pearson correlation coefficient = 0.847) between the segment index derived from apnea detection using the 1-dimensional convolutional neural network (1-D CNN) model and the AHI determined through PSG scoring (Fig. 2 A). In the Bland–Altman plots, the model's predicted AHI was generally lower than that measured by PSG for lower AHI values and higher for higher AHI values. The limits of agreement at the 95% CI were ± 20.525 (Fig. 2 B). Model versus sleep-related questionnaires for predicting moderate or higher OSA We compared the performance of our Model with that of commonly used sleep-related questionnaires for predicting moderate or high OSA. By determining the optimal cutoff scores using Youden's index, we found these to be ≥ 10 points for the PSQI and ≥ 7 points for the BDI-II. The points where Youden's index was maximized were confirmed at an ISI > 7 points, ESS > 14 points, and STOP-Bang > 4 points. Additionally, using a high-risk cutoff value, the Berlin questionnaire 14 , 15 was employed, and the results of the comparison, including accuracy, precision, recall, F1-score, and AUROC (95% CI), with the Model developed in this study are presented in Table 2 . Table 2 Performance comparison of sleep questionnaire and model in predicting moderate or severe obstructive sleep apnea (AHI ≥ 15) Variable Cut-off Accuracy Precision Recall F1-score AUROC (95% CI) PSQI ≥ 10 0.489 0.871 0.386 0.535 0.601 (0.464, 0.738) BDI-II ≥ 7 0.728 0.836 0.800 0.818 0.678 (0.540, 0.808) ISI ≥ 4 0.717 0.797 0.843 0.819 0.569 (0.431, 0.712) ESS ≥ 14 0.337 1.000 0.129 0.228 0.524 (0.388, 0.667) STOP-Bang ≥ 4 0.663 0.868 0.657 0.748 0.716 (0.581, 0.837) Berlin questionnaire High 0.685 0.500 0.379 0.431 0.621 (0.509, 0.728) Model 0.924 0.818 0.857 0.837 0.888 (0.793, 0.962) AUROC: area under the receiver operating characteristic; BDI-II: Beck Depression Inventory-II; CI: confidence interval; ESS: Epworth Sleepiness Scale; ISI: Insomnia Severity Index; PSQI: Pittsburgh Sleep Quality Index In evaluating the predictive performance for moderate or higher OSA, among various sleep-related questionnaires, STOP-Bang exhibited the highest AUROC at 0.716, followed by the BDI-II (AUROC 0.678), Berlin questionnaire (AUROC 0.621), and PSQI (AUROC 0.601). ISI (AUROC 0.569) and ESS (AUROC 0.524) demonstrated lower performance. The Model developed in this study showed robust results, with an accuracy of 0.924, precision of 0.818, recall of 0.857, F1-score of 0.837, and AUROC of 0.888 (95% CI: 0.793–0.962). Notably, the STOP-Bang exhibited the best performance among the sleep-related questionnaires. Compared with STOP-Bang, the Model demonstrated superior results in all aspects except precision. Confusion matrix analysis: Performance of the 1D-CNN Model in OSA severity prediction The confusion matrix demonstrated an overall moderate agreement (68.5% and kappa value of 0.526) between the classification of OSA using PSG and the 1-D CNN model (Fig. 4 ). In the assessment of OSA, 100% agreement was observed when predicting the absence of OSA. However, for predicting mild OSA, an agreement rate of 50% (7/14) was achieved. In cases where mild OSA was predicted, 28.6% (4/14) of the patients were confirmed to have no OSA, and an additional 21.4% (3/14) were identified as having moderate OSA. Conversely, when moderate OSA was predicted, 51.7% (15/29) of the cases were accurately categorized as moderate OSA, 13.8% (4/29) as mild OSA, and 34.5% (10/29) as severe OSA. Regarding the predictions of severe OSA, 81.0% (34/42) of the cases were correctly identified, while 19.0% (8/42) had moderate OSA. These findings highlighted the varying degrees of agreement and accuracy in predicting different levels of OSA severity. Discussion In this study, we utilized a 1D-CNN model constructed from EKG signals obtained via a wearable patch Holter monitoring device. The developed AI model demonstrated promising capabilities for accurately predicting the presence of sleep apnea in each analyzed segment. When assessing the Model's performance in predicting moderate or high OSA and comparing it with established clinical screening tools, our Model exhibited comparable predictive accuracy. We observed a moderate level of agreement between the OSA classification and the gold-standard PSG classification. Recent studies highlight the effectiveness of various AI models in sleep apnea detection with EKG signals 16 – 20 . The 1-D CNN model employed in this study offers the advantage of streamlining the signal preprocessing, particularly when applied to EKG signals using a wearable patch Holter device 21 . A previous study demonstrated promising results using a 1-D CNN model for sleep apnea detection from EKG signals using open databases 22 ; hence, this study demonstrated the potential of using a 1-D CNN model for sleep apnea detection using EKG signals from a wearable patch Holter device. The development of a model utilizing EKG signals from a clinical-grade Holter monitor, as opposed to open-dataset data or PSG EKG signals, enhances its applicability in real-world clinical settings. This approach addresses some limitations of PSG, such as single-night sleep measurements, by providing a more accessible and continuous monitoring solution. We compared the effectiveness of predicting moderate-to-severe OSA using various sleep questionnaires, emphasizing the importance of identifying such cases, as they warrant positive airway pressure treatment and appropriate management. Among the questionnaires, STOP-Bang displayed the highest predictive performance, which is consistent with prior research 23 , 24 . Our Model achieved an impressive accuracy rate of 92.4%, surpassing existing screening tools by more than 20%, highlighting its potential for accurately identifying individuals in need of treatment for OSA. The AI model demonstrated reasonable effectiveness in diagnosing and making treatment decisions for the AHI classification. However, its predictive accuracy varies across OSA severity levels. It was highly effective in identifying cases with no OSA (100.0%) and severe OSA (81.0%), but less effective in identifying mild (50.0%) and moderate (51.7%) cases. Our data suggest that when the AI predicts severe OSA, patients are likely to have moderate OSA. Therefore, prompt diagnosis and consideration of positive airway pressure are recommended. For the prediction of mild or moderate OSA, a more gradual approach involving repetitive prediction or the recommendation of PSG for confirmation may be appropriate. This differentiation is helpful in-patient management strategies. This study has several limitations. First, this was a single-center study, which prevented external validation. However, the research dataset encompassed diverse classifications of OSA in real-world scenarios, incorporating patients of various ages and sexes to enhance its clinical relevance. Second, while the wearable patch Holter device utilized in this study can detect respiration, our analysis focused solely on EKG signals. Although this approach enables a direct comparison of AI models with other studies emphasizing EKG signal detection, future research should explore the advantages of respiratory signal detection to develop a more robust OSA detection model. The present study had several strengths. First, it was conducted in patients clinically requiring PSG to ensure the necessity of screening. Among these patients, an 88.0% diagnostic rate of OSA was achieved, with 76.1% having moderate or high OSA requiring positive airway pressure treatment. This study emphasized the potential utility of a wearable patch Holter device for OSA screening in this patient population. Second, the detection methodology employed a segment-wise analysis, accounting for variations in breathing patterns at the night, based on wakefulness, sleep, sleep stage, and positive airway pressure device usage. This approach offers an understanding of changes in sleep status and pathophysiological states compared to measuring the AHI based on the entire night’s sleep. Third, we assessed the performance of the Model against commonly used clinical screening questionnaires for sleep apnea, demonstrating its clinical utility. Many clinical screening tools face limitations in clinical settings owing to their low accuracy, but the Model developed in this study exhibited high accuracy and precision, potentially overcoming existing screening limitations. In conclusion, this study demonstrates that an AI model developed from EKG signals captured by a wearable Holter monitoring device offers superior screening performance for OSA compared with previous screening questionnaires. These findings suggest its potential as a practical and accessible tool, particularly for patients facing challenges during PSG. Methods Data and participants The study participants who visited the Severance Hospital Sleep Clinic and underwent PSG between January and October 2022 were included. Inclusion criteria for adults aged 20 years and above included: 1) the presence of symptoms such as daytime sleepiness, frequent snoring, sleep apnea, fatigue, choking during sleep, frequent tossing and turning, and frequent awakening during sleep; 2) Modified Mallampatti score of grade 3 or higher, indicating difficulty during laryngotracheal intubation; and 3) presence of hypertension, heart disease, cerebrovascular disease, diabetes, or a body mass index (BMI) of 30 kg/m 2 or higher. Satisfying conditions 1) and 2) or conditions 1) and 3) were enough for inclusion. Pregnant women and patients with persistent atrial fibrillation, ventricular tachycardia, or implanted pacemakers were excluded. Informed consent was obtained from all eligible participants. This study was approved by the institutional review board of the Severance Hospital (No. 1-2021-0078). All the procedures were performed in accordance with the principles of the Declaration of Helsinki. Wearable patch Holter monitoring signal The HiCardi system, a wearable Holter EKG, facilitates extended monitoring of EKG and respiration in diverse patient groups. Comprised of the HiCardi terminal (Smart Patch), a mobile application (Smart View), and web viewer software (Live Studio), it provides a comprehensive clinical evaluation. The mobile application acts as a conduit for transmitting EKG data to a web server, and the web viewer aids the medical staff in data verification. By utilizing disposable electrodes, the device offers real-time EKG measurements accessible through both a smartphone application and a PC web viewer. To use the device, the protective film was peeled off from the disposable electrode, and the device was placed at the center of the patient's chest. The recommended attachment site is along the central sternal line, two-thirds of which are below the starting point. This device was affixed to the patients participating in the study concurrently with PSG, and the data were recorded (Fig. 3 A). Polysomnography We conducted overnight in-laboratory PSG recordings using Natus SleepWorks Software (Natus Medical Inc., USA). Patients were instructed to sleep in a controlled environment with dim lighting, temperature regulation, and noise control. The PSG recordings included electroencephalography with frontal, central, and occipital electrodes; 1-lead EKG; electromyography of extraocular eye movement, chin, and bilateral anterior tibialis muscles; nasal airflow and thermistor; peripheral oxygen saturation; sleep position; and chest and abdominal plethysmography. Sleep staging and scoring of respiratory events and movements followed the guidelines of the American Academy of Sleep Medicine (version 3.0) and were conducted by two sleep technicians with over 10 years of experience. Two neurologists (SHL and KMK), experienced in PSG interpretation, meticulously reviewed the PSG data. They reevaluated sleep staging and respiratory events, leading to the confirmation and approval of the final diagnosis and classification of OSA. Sleep questionnaires Questionnaires, including the Pittsburgh Sleep Quality Index (PSQI), Beck Depression Inventory-II (BDI-II), Insomnia Severity Index (ISI), Epworth Sleepiness Scale (ESS), STOP-Bang score, and Berlin questionnaire, were administered to assess the sleep habits and related problems of the participants before the PSG. The PSQI assesses sleep quality and disturbances on a scale ranging from 0–21, considering aspects such as sleep duration, efficiency, interruptions, and daytime dysfunction over a 1-month period 25 . BDI-II is a 21-item self-report questionnaire that assesses the severity of depressive symptoms. Respondents rated their statements based on their experiences over the past 2 weeks. The total score ranged from 0–63, with higher scores indicating more severe depressive symptoms 26 . The ISI is a seven-item questionnaire measuring insomnia severity over the past 2 weeks, with scores ranging from 0–28. It assesses various aspects of sleep disruption and its impact on daily life 27 . The ESS is an eight-item questionnaire evaluating daytime sleepiness. Participants rated their likelihood of dozing off in different situations, with scores ranging from 0–24 28 . The STOP-Bang questionnaire scores individuals on eight factors ranging from 0–8, including snoring, tiredness, observed apnea, high blood pressure, BMI, age (≥ 50 years), neck circumference, and male sex 14 , 15 , 29 . Higher scores indicate a higher risk of OSA. The Berlin questionnaire evaluates the risk of sleep apnea through snoring, daytime fatigue, and hypertension/obesity. It categorizes individuals as high- or low- risk 30 . Data Pre-processing The EKG signal collected from the wearable patch Holter device may experience some packet loss as it is transmitted over the wireless network. Therefore, the signal is compensated for by filling in the intervals where packet loss occurs with zeros so that the signal is continuous. The corrected signal uses an adaptive notch filter to remove the 60Hz band to remove powerline noise, and a high pass filter of 0.5Hz and a low pass filter of 40Hz for smooth analysis. An additional adaptive filter is applied to reduce baseline wander. Deep learning We developed a 1-dimensional dilated convolutional neural network utilizing a 3-minute EKG signal as an input to predict sleep apnea (Fig. 3 B). All recordings were divided into 3-minute segments. The training and test sets were assigned by randomly shuffling all segments in a 7:3 ratio. Our Model exclusively uses raw EKG signals that have undergone z-score normalization without relying on features derived from EKG or the demographic data of the subjects. The model comprised of four convolutional blocks and an inference block. The convolutional blocks were responsible for identifying and compressing features from the EKG signals for sleep apnea detection, whereas the inference block predicted the presence or absence of sleep apnea based on the extracted features. Each convolutional block consisted of a dilated convolutional layer and a max-pooling layer, whereas the inference block included a Global Average Pooling (GAP) layer and a dense layer. Thus, the Model used comprised of ten layers. Considering the specifics of EKG signals, the Model is based on 1-D convolutional layers, but utilizes dilated convolutional layers instead of standard convolutional layers 31 . Dilated convolution accommodates a broader receptive field by adjusting the spacing between the filters, which is a feature controlled by the dilation rate. A dilation rate of 1 equates to standard convolution, whereas a rate of 2 introduces zero-filled spaces between filter elements, allowing broader coverage of input values with the same filter size and a deeper understanding of the input data. Due to these characteristics, each filter in the model extracts features that are effective in distinguishing between normal and apnea/hypopnea. We experimented with various dilation rates from 1 to 6 and ultimately set the dilation rate to 2. A small dilation rate captured a wider area and showed improved results; whereas, a large dilation rate was expected to fail in capturing important features owing to the spacing between filters. Each dilated convolutional layer employed a filter size of 10, with a total of 200 filters and a stride size of 1. Padding was performed to ensure equal input and output sizes of the layers. The feature maps generated from the dilated convolutional layer were processed through a rectified linear unit activation function before being input into the max-pooling layer. The pool size in the maximum pooling layer is 2, halving the size of the input feature map. The principal features selected by the max-pooling layer are then transferred to the next convolutional block. The features passing through the four convolutional blocks were fed into the GAP layer of the inference block, converting them into a 1-dimensional array corresponding to the size of the feature map extracted from the convolutional blocks. Each element of the array represents the average value at that location on the feature map. GAP reduces the dimensions by averaging the feature map, significantly lowering the Model’s parameter count compared to a flattened layer, thereby reducing the overfitting risk. Because GAP averages the entire feature map, it preserves information irrespective of the specific locations of the main features in each map. Finally, a 1-dimensional array of features processed through GAP was utilized to output prediction values via a dense layer. The batch size is set to 64 to maximize the memory of the A100 GPU during the model training process. One of the hyper-parameters, epoch, is initially set to 100. Epoch is set to automatically terminate training without further epoch progression if the model is suspected to be overfitting by the early stopping method during the training process. As a result, training was terminated at 94 epochs by the early stopping method. The supplementary information figure S1 shows the respective loss and accuracy values for train and validation for each epoch. Also, some of the key hyper-parameters, optimizer and learning rate, are set to 'Nadam' optimizer and 0.001 respectively. This was chosen after experimenting with several optimizers and learning rate ratios, and the supplementary information table S1 shows the experimental results for each optimizer and learning rate. Table S1 shows training loss, training accuracy, and validation accuracy respectively, where epoch in the table refers to the point at which training was terminated by the early stopping method. We stop learning when the validation accuracy reaches a peak and the validation accuracy does not increase after about 10 epochs. In the optimizer comparison experiment, we fixed the learning rate to 0.001, and in the learning rate comparison experiment, we fixed Nadam, the best performer in the optimizer comparison experiment, as the optimizer. Assessment and comparison of the Model, questionnaires, and PSG classification The segment index was defined as the ratio of the segments predicted by the final Model to be apnea segments to the total number of segments for each participant. We used Pearson’s correlation analysis to evaluate the correlation between the Model's segment index and the PSG-derived apnea-hypopnea index (AHI) 32 , 33 . The AHI predicted by the Model was obtained by applying the segment index to Pearson's correlation analysis. The variability between the Model-predicted AHI and the AHI derived from PSG was illustrated using Bland–Altman plots. In addition, for moderate or higher OSA prediction, cutoff values for sleep-related questionnaires were determined using Youden's index, calculated as sensitivity plus specificity minus one, with a range from 0 to 1 34 . Subsequently, we compared the Model's performance using sleep-related questionnaires based on the accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUROC) (95% confidence interval [CI]). The accuracy of the Model in classifying OSA was demonstrated using a confusion matrix with PSG, and the Cohen’s kappa coefficient was used to evaluate the agreement between PSG and AI models in classifying OSA 16 , 35 . All statistical analyses were performed using Python version 3.12.1. Statistical significance was set at a two-tailed p < 0.050. Declarations Conflict of Interest statement All authors have no conflicts to disclose. Funding This research was supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: HI20C2125). Author Contribution SHL and MK: Acquisition of data, data analysis and interpretation, and manuscript writing. SPC, YK, MKC, W-JK, and KH: Revision of manuscript critically for important intellectual content. KMK: Conception and design, data analysis and interpretation, supervision, and final approval of the version to be submitted. Acknowledgements None Data Availability Anonymized data relevant to this study will be shared upon request with a qualified investigator pending appropriate Institutional Review Board approvals. References Kapur, V. K. & Weaver, E. M. Filling in the pieces of the sleep apnea-hypertension puzzle. Jama 307 , 2197–2198. https://doi.org:10.1001/jama.2012.5039 (2012). Gami, A. S. et al. Obstructive sleep apnea and the risk of sudden cardiac death: a longitudinal study of 10,701 adults. J. Am. Coll. Cardiol. 62 , 610–616. https://doi.org:10.1016/j.jacc.2013.04.080 (2013). Durgan, D. J. & Bryan, R. M. Jr. Cerebrovascular consequences of obstructive sleep apnea. J. Am. Heart Assoc. 1 , e000091. https://doi.org:10.1161/jaha.111.000091 (2012). Reichmuth, K. J., Austin, D., Skatrud, J. B. & Young, T. Association of sleep apnea and type II diabetes: a population-based study. Am. J. Respir Crit. Care Med. 172 , 1590–1595. https://doi.org:10.1164/rccm.200504-637OC (2005). Engleman, H. M., Douglas, N. J. & Sleep 4: Sleepiness, cognitive function, and quality of life in obstructive sleep apnoea/hypopnoea syndrome. Thorax 59, 618–622 (2004). https://doi.org:10.1136/thx.2003.015867 Benjafield, A. V. et al. Estimation of the global prevalence and burden of obstructive sleep apnoea: a literature-based analysis. Lancet Respir Med. 7 , 687–698. https://doi.org:10.1016/s2213-2600(19)30198-5 (2019). Zou, J. et al. An effective model for screening obstructive sleep apnea: a large-scale diagnostic study. PLoS One . 8 , e80704. https://doi.org:10.1371/journal.pone.0080704 (2013). Antic, N. A. et al. A randomized controlled trial of nurse-led care for symptomatic moderate-severe obstructive sleep apnea. Am. J. Respir Crit. Care Med. 179 , 501–508. https://doi.org:10.1164/rccm.200810-1558OC (2009). Böhning, N. et al. Comparability of pulse oximeters used in sleep medicine for the screening of OSA. Physiol. Meas. 31 , 875–888. https://doi.org:10.1088/0967-3334/31/7/001 (2010). Dawson, A. et al. Type III home sleep testing versus pulse oximetry: is the respiratory disturbance index better than the oxygen desaturation index to predict the apnoea-hypopnoea index measured during laboratory polysomnography? BMJ Open. 5 , e007956. https://doi.org:10.1136/bmjopen-2015-007956 (2015). Bazoukis, G. et al. Application of artificial intelligence in the diagnosis of sleep apnea. J. Clin. Sleep. Med. 19 , 1337–1363. https://doi.org:10.5664/jcsm.10532 (2023). Murali, S. et al. Cardiac Ambulatory Monitoring: New Wireless Device Validated Against Conventional Holter Monitoring in a Case Series. Front. Cardiovasc. Med. 7 , 587945. https://doi.org:10.3389/fcvm.2020.587945 (2020). Duncker, D. et al. Smart Wearables for Cardiac Monitoring-Real-World Use beyond Atrial Fibrillation. Sens. (Basel) . 21. https://doi.org:10.3390/s21072539 (2021). Nagappa, M. 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 . 10 , e0143697. https://doi.org:10.1371/journal.pone.0143697 (2015). Jeon, H. J., Bang, Y. R. & Yoon, I. Y. A validation study on three screening questionnaires for obstructive sleep apnea in a Korean community sample. Sleep. Breath. 23 , 969–977. https://doi.org:10.1007/s11325-018-1748-x (2019). Cohen, J. A coefficient of agreement for nominal scales. Educ. Psychol. Meas. 20 , 37–46 (1960). Salari, N. et al. Detection of sleep apnea using Machine learning algorithms based on ECG Signals: A comprehensive systematic review. Expert Syst. Appl. 187 , 115950 (2022). Erdenebayar, U., Kim, Y. J., Park, J. U., Joo, E. Y. & Lee, K. J. Deep learning approaches for automatic detection of sleep apnea events from an electrocardiogram. Comput. Methods Programs Biomed. 180 , 105001. https://doi.org:10.1016/j.cmpb.2019.105001 (2019). Hassan, A. R. & Haque, M. A. Computer-aided obstructive sleep apnea screening from single-lead electrocardiogram using statistical and spectral features and bootstrap aggregating. Biocybernetics Biomedical Eng. 36 , 256–266 (2016). Babaeizadeh, S., White, D. P., Pittman, S. D. & Zhou, S. H. Automatic detection and quantification of sleep apnea using heart rate variability. J. Electrocardiol. 43 , 535–541. https://doi.org:10.1016/j.jelectrocard.2010.07.003 (2010). Singstad, B. J. & Tronstad, C. in 2020 computing in cardiology. 1–4 (IEEE). Chang, H. Y., Yeh, C. Y., Lee, C. T. & Lin, C. C. A Sleep Apnea Detection System Based on a One-Dimensional Deep Convolution Neural Network Model Using Single-Lead Electrocardiogram. Sens. (Basel) . 20. https://doi.org:10.3390/s20154157 (2020). Chung, F., Abdullah, H. R., Liao, P. & STOP-Bang Questionnaire A Practical Approach to Screen for Obstructive Sleep Apnea. Chest 149 , 631–638. https://doi.org:10.1378/chest.15-0903 (2016). Tan, A. et al. Predicting obstructive sleep apnea using the STOP-Bang questionnaire in the general population. Sleep. Med. 27–28 , 66–71. https://doi.org:10.1016/j.sleep.2016.06.034 (2016). Buysse, D. J., Reynolds, C. F. 3, Monk, T. H., Berman, S. R., Kupfer, D. J. & rd, & The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res. 28 , 193–213. https://doi.org:10.1016/0165-1781(89)90047-4 (1989). Beck, A. T., Steer, R. A., Ball, R. & Ranieri, W. Comparison of Beck Depression Inventories -IA and -II in psychiatric outpatients. J. Pers. Assess. 67 , 588–597. https://doi.org:10.1207/s15327752jpa6703_13 (1996). Morin, C. M., Belleville, G., Bélanger, L. & Ivers, H. The Insomnia Severity Index: psychometric indicators to detect insomnia cases and evaluate treatment response. Sleep 34 , 601–608. https://doi.org:10.1093/sleep/34.5.601 (2011). Doneh, B. Epworth Sleepiness Scale. Occup. Med. (Lond) . 65 , 508. https://doi.org:10.1093/occmed/kqv042 (2015). Hwang, M. et al. Validation of the STOP-Bang questionnaire as a preoperative screening tool for obstructive sleep apnea: a systematic review and meta-analysis. BMC Anesthesiol . 22 , 366. https://doi.org:10.1186/s12871-022-01912-1 (2022). Netzer, N. C., Stoohs, R. A., Netzer, C. M., Clark, K. & Strohl, K. P. Using the Berlin Questionnaire to identify patients at risk for the sleep apnea syndrome. Ann. Intern. Med. 131 , 485–491. https://doi.org:10.7326/0003-4819-131-7-199910050-00002 (1999). Yu, F. & Koltun, V. Multi-scale context aggregation by dilated convolutions. arXiv preprint arXiv:1511.07122 (2015). Stigler, S. M. Francis Galton's account of the invention of correlation. Statistical Science , 73–79 (1989). Bland, J. M. & Altman, D. G. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1 , 307–310 (1986). Youden, W. J. Index for rating diagnostic tests. Cancer 3 , 32–35. https://doi.org:10.1002/1097-0142(1950)3:1<32::aid-cncr2820030106>3.0.co;2-3 (1950). Stehman, S. V. Selecting and interpreting measures of thematic classification accuracy. Remote Sens. Environ. 62 , 77–89 (1997). Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterials.docx Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 20 Oct, 2025 Reviews received at journal 19 Oct, 2025 Reviews received at journal 17 Oct, 2025 Reviews received at journal 14 Oct, 2025 Reviewers agreed at journal 06 Oct, 2025 Reviewers agreed at journal 03 Oct, 2025 Reviewers agreed at journal 02 Oct, 2025 Reviewers agreed at journal 01 Oct, 2025 Reviewers agreed at journal 01 Oct, 2025 Reviewers agreed at journal 01 Oct, 2025 Reviewers invited by journal 09 Sep, 2025 Editor invited by journal 09 Sep, 2025 Editor assigned by journal 08 Sep, 2025 Submission checks completed at journal 06 Sep, 2025 First submitted to journal 04 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-7537553\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Article\",\"associatedPublications\":[],\"authors\":[{\"id\":514853775,\"identity\":\"ca1d65f6-a7e1-4849-a4b0-b08fc3333091\",\"order_by\":0,\"name\":\"Sue Hyun Lee\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"National Health Insurance Service Ilsan Hospital\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Sue\",\"middleName\":\"Hyun\",\"lastName\":\"Lee\",\"suffix\":\"\"},{\"id\":514853777,\"identity\":\"49dfa76c-d592-4a7f-93e8-4ba898ddb4c3\",\"order_by\":1,\"name\":\"Minwoo Kim\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"MEZOO Co., Ltd\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Minwoo\",\"middleName\":\"\",\"lastName\":\"Kim\",\"suffix\":\"\"},{\"id\":514853778,\"identity\":\"67c8bbfa-83af-4bc9-880e-e42fc8cf7476\",\"order_by\":2,\"name\":\"Sung Pil Cho\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"MEZOO Co., Ltd\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Sung\",\"middleName\":\"Pil\",\"lastName\":\"Cho\",\"suffix\":\"\"},{\"id\":514853779,\"identity\":\"355c7ab0-20f1-469b-bcca-45c18739dc16\",\"order_by\":3,\"name\":\"Yeewoong Kim\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"MEZOO Co., Ltd\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Yeewoong\",\"middleName\":\"\",\"lastName\":\"Kim\",\"suffix\":\"\"},{\"id\":514853781,\"identity\":\"d798a55a-834c-4e0f-a1ab-236a181e09b5\",\"order_by\":4,\"name\":\"Min Kyung Chu\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Yonsei University College of Medicine\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Min\",\"middleName\":\"Kyung\",\"lastName\":\"Chu\",\"suffix\":\"\"},{\"id\":514853782,\"identity\":\"898b91f1-8cc1-4a08-9bf2-e2cfaf2bfc44\",\"order_by\":5,\"name\":\"Won-Joo Kim\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Yonsei University College of Medicine\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Won-Joo\",\"middleName\":\"\",\"lastName\":\"Kim\",\"suffix\":\"\"},{\"id\":514853783,\"identity\":\"132589f5-1db0-4916-b4cc-ac6bbbca37f8\",\"order_by\":6,\"name\":\"Kyoung Heo\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Yonsei University College of Medicine\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Kyoung\",\"middleName\":\"\",\"lastName\":\"Heo\",\"suffix\":\"\"},{\"id\":514853784,\"identity\":\"af2f1f4c-d1aa-4e78-92ce-3ef6c8966d7d\",\"order_by\":7,\"name\":\"Kyung Min Kim\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAElEQVRIiWNgGAWjYHACxgMJDAcY2ICsAwkVDAwGxOgBaZFgA+l9cIZYLUAkAaSZDz5sI0KLOXvvgQMPft2p45Nuv3Agcd7hxO38Bxg//MCjxbLnXMKBxL5nEmwyZwoOJG47nLhzRgKzZA8eLQY3cgwOJPYclmCTyEkAa9lwg4FBGp/DDO6/QdYyB6jl/AHm33i13OAxOJDwA6Ql/cCBxAaglgMJbHhtsewBOazhsGSbRA4wtI+lG2+4kdhmic8v5uxnDB/++HOYX35G+uOPP2qsZTecP3z4Br4QA8cCYxuI5AGxm0HcBnzugkbcHxDB/gBI1OFVPQpGwSgYBSMTAABrSGIbqi52XgAAAABJRU5ErkJggg==\",\"orcid\":\"\",\"institution\":\"Yonsei University College of Medicine\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Kyung\",\"middleName\":\"Min\",\"lastName\":\"Kim\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-09-04 15:23:15\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-7537553/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-7537553/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":91560391,\"identity\":\"7ed8cede-8ecf-4858-8283-30a314df4a5d\",\"added_by\":\"auto\",\"created_at\":\"2025-09-17 18:42:12\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":454097,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eModel performance in apnea detection\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7537553/v1/8a1341e271ca53eb9c3563d1.png\"},{\"id\":91560397,\"identity\":\"621f32ba-55e7-431c-a726-bbfe0355c980\",\"added_by\":\"auto\",\"created_at\":\"2025-09-17 18:42:13\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1024951,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e(A) Correlation graph and (B) Bland-Altman plots between segment index obtained by the Model and apnea-hypopnea index of polysomnography\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure2AB.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7537553/v1/439cb10299bdd67a6d7ec488.png\"},{\"id\":91560393,\"identity\":\"c43a5bfd-1b8c-45d8-a801-2a89fd530fc0\",\"added_by\":\"auto\",\"created_at\":\"2025-09-17 18:42:13\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1432969,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003e(A) Wearable patch Holter monitoring device and (B) 1-dimensional convolutional neural network algorithm\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure3AB.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7537553/v1/10b4b27dc5d2f55fe1bd2825.png\"},{\"id\":91563244,\"identity\":\"d10dd9aa-776c-4161-9a20-e0a75ea79db6\",\"added_by\":\"auto\",\"created_at\":\"2025-09-17 19:06:13\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":323566,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eThe confusion matrix compares the classification of obstructive sleep apnea (OSA) diagnosed using polysomnography and a 1-D convolutional neural network model. The matrix illustrates the division into the following four categories: no OSA, mild OSA, moderate OSA, and severe OSA.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7537553/v1/1e3739eab1e7889cdadbe54f.png\"},{\"id\":91817132,\"identity\":\"2dbd5ff2-9a04-4198-bfe4-10004e056c42\",\"added_by\":\"auto\",\"created_at\":\"2025-09-22 06:53:40\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":3836489,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7537553/v1/98b8a6b8-0d9a-4e4b-9bf4-5edb8e8a6c78.pdf\"},{\"id\":91560399,\"identity\":\"4a3d5345-81bc-4192-b429-dd133697686c\",\"added_by\":\"auto\",\"created_at\":\"2025-09-17 18:42:13\",\"extension\":\"docx\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":216913,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementarymaterials.docx\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7537553/v1/cfedd6b453e4e3a9bcb3bd0d.docx\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Sleep Apnea Detection Using Wearable ECG and Deep Learning: Validation with Polysomnography\",\"fulltext\":[{\"header\":\"Introduction\",\"content\":\"\\u003cp\\u003eObstructive sleep apnea (OSA) is a common sleep-related breathing disorder characterized by repeated upper airway obstruction during sleep, resulting in intermittent hypoxia and sleep fragmentation. It has been associated with a wide range of adverse health outcomes, including hypertension, cardiovascular and cerebrovascular diseases, metabolic dysfunction, and impaired daytime functioning\\u003csup\\u003e\\u003cspan additionalcitationids=\\\"CR2 CR3 CR4\\\" citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e5\\u003c/span\\u003e\\u003c/sup\\u003e. Epidemiological studies estimate that approximately 936\\u0026nbsp;million adults worldwide are affected by mild-to-severe OSA, and over 400\\u0026nbsp;million individuals suffer from moderate-to-severe forms, underscoring its significant public health impact\\u003csup\\u003e\\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e\\u003c/sup\\u003e.\\u003c/p\\u003e\\u003cp\\u003eDespite its prevalence, OSA remains underdiagnosed due to limitations in current diagnostic methods. Polysomnography (PSG), the clinical gold standard, is resource-intensive, time-consuming, and typically restricted to specialized sleep laboratories. Alternative approaches such as home sleep apnea testing and pulse oximetry offer greater accessibility but are often limited by reduced sensitivity and challenges in signal quality during natural sleep movement\\u003csup\\u003e\\u003cspan additionalcitationids=\\\"CR8 CR9\\\" citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e\\u003c/sup\\u003e.\\u003c/p\\u003e\\u003cp\\u003eRecent advances in artificial intelligence (AI) and wearable biosensors have introduced new possibilities for OSA detection using physiological signals such as electrocardiograms (ECG)\\u003csup\\u003e\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e\\u003c/sup\\u003e. Specifically, deep learning models have shown promising performance in extracting apnea-relevant features directly from ECG signals, without the need for manual feature engineering. Wearable patch-type ECG devices further enhance the feasibility of out-of-hospital sleep monitoring due to their simplicity, comfort, and continuous data acquisition\\u003csup\\u003e\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e\\u003c/sup\\u003e.\\u003c/p\\u003e\\u003cp\\u003eIn this study, we developed and validated a deep learning model for sleep apnea detection using ECG signals collected from a wearable patch Holter device. The model was trained to classify apnea events based on 3-minute ECG segments and evaluated against PSG-derived apnea-hypopnea index (AHI) scores. Our findings suggest that this approach may offer a scalable, non-invasive solution for OSA screening, with potential applications in both clinical and community settings.\\u003c/p\\u003e\"},{\"header\":\"Results\",\"content\":\"\\u003cdiv id=\\\"Sec3\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eClinical characteristics\\u003c/h2\\u003e\\u003cp\\u003eThe demographic and clinical characteristics of the study participants are shown in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e. We recruited a cohort of 92 patients with an average age of 55.1 years. Men constituted a significant majority of the patients (75.0%, 69 patients). Our study population exhibited 26.0\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;4.4 kg/m\\u003csup\\u003e2\\u003c/sup\\u003e for BMI, indicating an overweight status. In terms of OSA severity, the distribution was as follows: no OSA was observed in 12.0% of the patients (11 patients), while mild OSA was diagnosed in 12.0% (11 patients), moderate OSA in 28.3% (26 patients), and severe OSA in 47.8% (44 patients) of patients.\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eDemographic and Clinical Characteristics of Patients in the Study\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"2\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eCharacteristics\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eSubjects, n\\u0026thinsp;=\\u0026thinsp;92\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAge (year)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e55.1\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;14.7\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eGender (male), n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e69 (75%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eBody mass index (kg/m\\u003csup\\u003e2\\u003c/sup\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e26.0\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;4.4\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003ePSQI\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e8.0 (5\\u0026ndash;11)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eISI\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e9.0 (4\\u0026ndash;14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eESS\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e6.0 (4\\u0026ndash;10)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eSTOP-BANG\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e4.0 (3\\u0026ndash;5)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eBDI-II\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e10 (6\\u0026ndash;18)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTime in bed (hours)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e7.4\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;0.8\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTotal sleep time (hours)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e5.7\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;1.2\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eSleep efficiency (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e76.9\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;17.3\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eArousal index (events/hour)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e37.5\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;18.8\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eO2 saturation nadir (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e83.9\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;7.5\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eApnea-hypopnea index (events/hour)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e32.9\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;23.4\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eOSA severity\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eNo OSA, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e11 (12.0%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eMild OSA, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e11 (12.0%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eModerate OSA, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e26 (28.3%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eSevere OSA, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e44 (47.8%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003ctfoot\\u003e\\u003ctr\\u003e\\u003ctd colspan=\\\"2\\\"\\u003eData are presented as mean\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;standard deviation for normally distributed quantitative data. For non-normally distributed quantitative data, data are presented as the medians with the interquartile range (25th\\u0026ndash;75th percentiles). BDI-II: Beck Depression Inventory-II; ESS: Epworth Sleepiness Scale; ISI: Insomnia Severity Index; OSA: Obstructive Sleep Apnea; PSQI: Pittsburgh Sleep Quality Index.\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tfoot\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eMedian values and interquartile ranges were calculated for the questionnaires. The median PSQI, ISI, ESS, and STOP-Bang scores were 8.0, 9.0, 6.0, and 4.0, respectively. These values indicate that our study population tended to have poor sleep quality, mild insomnia, normal levels of daytime sleepiness, and an intermediate risk of OSA. The median BDI-II score was 10.0, indicating that the patients were less likely to have depression.\\u003c/p\\u003e\\u003cp\\u003eThe mean value of time in bed was 7.4\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;0.8 hours, and the total sleep time was 5.7\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;1.2 hours. The sleep efficiency was 76.9\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;17.3%, showing decreased sleep efficiency. However, this study was conducted in a hospital for overnight laboratory PSG recordings, which may have affected sleep efficiency. The arousal index was 37.5\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;18.8 events/hour, which is a markedly increased value, even considering the first-night effect. The mean AHI was 32.9\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;23.4 events/ hour, and the mean O2 saturation nadir was 83.9\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;7.5%.\\u003c/p\\u003e\\u003c/div\\u003e\\n\\u003ch3\\u003eModel performance in apnea detection\\u003c/h3\\u003e\\n\\u003cp\\u003eWe experimented with segments of different lengths based on the model configuration described earlier. We experimented with segments of 5 minutes, 3 minutes, 1 minute, and 30 seconds, and found that the 3-minute segment performed the best, as shown in the supplementary information table S2. The results of our experiments using the 3-minute segment are as follows. The total number of segments was 13,916, wherein 9,741 segments were used as the training set and 4,175 segments were used as the test set. The 9,741 training sets consisted of 4,649 normal segments and 5,092 apnea segments, whereas the 4,175 test sets consisted of 1,965 normal segments and 2,210 apnea segments. The final model showed an accuracy of 0.811, precision of 0.819, recall of 0.825, F1 score of 0.822, and AUROC of 0.875 (95% CI, 0.864\\u0026ndash;0.886) when evaluating the apnea detection performance on the test set (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\n\\u003ch3\\u003eCorrelation and Bland-Altman Plots\\u003c/h3\\u003e\\n\\u003cp\\u003eWe observed a strong linear correlation (Pearson correlation coefficient\\u0026thinsp;=\\u0026thinsp;0.847) between the segment index derived from apnea detection using the 1-dimensional convolutional neural network (1-D CNN) model and the AHI determined through PSG scoring (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003eA). In the Bland\\u0026ndash;Altman plots, the model's predicted AHI was generally lower than that measured by PSG for lower AHI values and higher for higher AHI values. The limits of agreement at the 95% CI were \\u0026plusmn;\\u0026thinsp;20.525 (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003eB).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\n\\u003ch3\\u003eModel versus sleep-related questionnaires for predicting moderate or higher OSA\\u003c/h3\\u003e\\n\\u003cp\\u003eWe compared the performance of our Model with that of commonly used sleep-related questionnaires for predicting moderate or high OSA. By determining the optimal cutoff scores using Youden's index, we found these to be \\u0026ge;\\u0026thinsp;10 points for the PSQI and \\u0026ge;\\u0026thinsp;7 points for the BDI-II. The points where Youden's index was maximized were confirmed at an ISI\\u0026thinsp;\\u0026gt;\\u0026thinsp;7 points, ESS\\u0026thinsp;\\u0026gt;\\u0026thinsp;14 points, and STOP-Bang\\u0026thinsp;\\u0026gt;\\u0026thinsp;4 points. Additionally, using a high-risk cutoff value, the Berlin questionnaire\\u003csup\\u003e\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e\\u003c/sup\\u003e was employed, and the results of the comparison, including accuracy, precision, recall, F1-score, and AUROC (95% CI), with the Model developed in this study are presented in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e.\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003ePerformance comparison of sleep questionnaire and model in predicting moderate or severe obstructive sleep apnea (AHI\\u0026thinsp;\\u0026ge;\\u0026thinsp;15)\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"7\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eVariable\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eCut-off\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eAccuracy\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003ePrecision\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eRecall\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003eF1-score\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003eAUROC (95% CI)\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003ePSQI\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u0026ge;\\u0026thinsp;10\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.489\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.871\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.386\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.535\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.601 (0.464, 0.738)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eBDI-II\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u0026ge;\\u0026thinsp;7\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.728\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.836\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.800\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.818\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.678 (0.540, 0.808)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eISI\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u0026ge;\\u0026thinsp;4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.717\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.797\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.843\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.819\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.569 (0.431, 0.712)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eESS\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u0026ge;\\u0026thinsp;14\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.337\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e1.000\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.129\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.228\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.524 (0.388, 0.667)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eSTOP-Bang\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u0026ge;\\u0026thinsp;4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.663\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.868\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.657\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.748\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.716 (0.581, 0.837)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eBerlin questionnaire\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eHigh\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.685\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.500\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.379\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.431\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.621 (0.509, 0.728)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eModel\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.924\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.818\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0.857\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.837\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.888 (0.793, 0.962)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003ctfoot\\u003e\\u003ctr\\u003e\\u003ctd colspan=\\\"7\\\"\\u003eAUROC: area under the receiver operating characteristic; BDI-II: Beck Depression Inventory-II; CI: confidence interval; ESS: Epworth Sleepiness Scale; ISI: Insomnia Severity Index; PSQI: Pittsburgh Sleep Quality Index\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tfoot\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eIn evaluating the predictive performance for moderate or higher OSA, among various sleep-related questionnaires, STOP-Bang exhibited the highest AUROC at 0.716, followed by the BDI-II (AUROC 0.678), Berlin questionnaire (AUROC 0.621), and PSQI (AUROC 0.601). ISI (AUROC 0.569) and ESS (AUROC 0.524) demonstrated lower performance. The Model developed in this study showed robust results, with an accuracy of 0.924, precision of 0.818, recall of 0.857, F1-score of 0.837, and AUROC of 0.888 (95% CI: 0.793\\u0026ndash;0.962). Notably, the STOP-Bang exhibited the best performance among the sleep-related questionnaires. Compared with STOP-Bang, the Model demonstrated superior results in all aspects except precision.\\u003c/p\\u003e\\n\\u003ch3\\u003eConfusion matrix analysis: Performance of the 1D-CNN Model in OSA severity prediction\\u003c/h3\\u003e\\n\\u003cp\\u003eThe confusion matrix demonstrated an overall moderate agreement (68.5% and kappa value of 0.526) between the classification of OSA using PSG and the 1-D CNN model (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e). In the assessment of OSA, 100% agreement was observed when predicting the absence of OSA. However, for predicting mild OSA, an agreement rate of 50% (7/14) was achieved. In cases where mild OSA was predicted, 28.6% (4/14) of the patients were confirmed to have no OSA, and an additional 21.4% (3/14) were identified as having moderate OSA. Conversely, when moderate OSA was predicted, 51.7% (15/29) of the cases were accurately categorized as moderate OSA, 13.8% (4/29) as mild OSA, and 34.5% (10/29) as severe OSA. Regarding the predictions of severe OSA, 81.0% (34/42) of the cases were correctly identified, while 19.0% (8/42) had moderate OSA. These findings highlighted the varying degrees of agreement and accuracy in predicting different levels of OSA severity.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\"},{\"header\":\"Discussion\",\"content\":\"\\u003cp\\u003eIn this study, we utilized a 1D-CNN model constructed from EKG signals obtained via a wearable patch Holter monitoring device. The developed AI model demonstrated promising capabilities for accurately predicting the presence of sleep apnea in each analyzed segment. When assessing the Model's performance in predicting moderate or high OSA and comparing it with established clinical screening tools, our Model exhibited comparable predictive accuracy. We observed a moderate level of agreement between the OSA classification and the gold-standard PSG classification.\\u003c/p\\u003e\\u003cp\\u003eRecent studies highlight the effectiveness of various AI models in sleep apnea detection with EKG signals\\u003csup\\u003e\\u003cspan additionalcitationids=\\\"CR17 CR18 CR19\\\" citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e16\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e20\\u003c/span\\u003e\\u003c/sup\\u003e. The 1-D CNN model employed in this study offers the advantage of streamlining the signal preprocessing, particularly when applied to EKG signals using a wearable patch Holter device\\u003csup\\u003e\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e\\u003c/sup\\u003e. A previous study demonstrated promising results using a 1-D CNN model for sleep apnea detection from EKG signals using open databases\\u003csup\\u003e\\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e\\u003c/sup\\u003e; hence, this study demonstrated the potential of using a 1-D CNN model for sleep apnea detection using EKG signals from a wearable patch Holter device. The development of a model utilizing EKG signals from a clinical-grade Holter monitor, as opposed to open-dataset data or PSG EKG signals, enhances its applicability in real-world clinical settings. This approach addresses some limitations of PSG, such as single-night sleep measurements, by providing a more accessible and continuous monitoring solution.\\u003c/p\\u003e\\u003cp\\u003eWe compared the effectiveness of predicting moderate-to-severe OSA using various sleep questionnaires, emphasizing the importance of identifying such cases, as they warrant positive airway pressure treatment and appropriate management. Among the questionnaires, STOP-Bang displayed the highest predictive performance, which is consistent with prior research\\u003csup\\u003e\\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e23\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR24\\\" class=\\\"CitationRef\\\"\\u003e24\\u003c/span\\u003e\\u003c/sup\\u003e. Our Model achieved an impressive accuracy rate of 92.4%, surpassing existing screening tools by more than 20%, highlighting its potential for accurately identifying individuals in need of treatment for OSA.\\u003c/p\\u003e\\u003cp\\u003eThe AI model demonstrated reasonable effectiveness in diagnosing and making treatment decisions for the AHI classification. However, its predictive accuracy varies across OSA severity levels. It was highly effective in identifying cases with no OSA (100.0%) and severe OSA (81.0%), but less effective in identifying mild (50.0%) and moderate (51.7%) cases. Our data suggest that when the AI predicts severe OSA, patients are likely to have moderate OSA. Therefore, prompt diagnosis and consideration of positive airway pressure are recommended. For the prediction of mild or moderate OSA, a more gradual approach involving repetitive prediction or the recommendation of PSG for confirmation may be appropriate. This differentiation is helpful in-patient management strategies.\\u003c/p\\u003e\\u003cp\\u003eThis study has several limitations. First, this was a single-center study, which prevented external validation. However, the research dataset encompassed diverse classifications of OSA in real-world scenarios, incorporating patients of various ages and sexes to enhance its clinical relevance. Second, while the wearable patch Holter device utilized in this study can detect respiration, our analysis focused solely on EKG signals. Although this approach enables a direct comparison of AI models with other studies emphasizing EKG signal detection, future research should explore the advantages of respiratory signal detection to develop a more robust OSA detection model.\\u003c/p\\u003e\\u003cp\\u003eThe present study had several strengths. First, it was conducted in patients clinically requiring PSG to ensure the necessity of screening. Among these patients, an 88.0% diagnostic rate of OSA was achieved, with 76.1% having moderate or high OSA requiring positive airway pressure treatment. This study emphasized the potential utility of a wearable patch Holter device for OSA screening in this patient population. Second, the detection methodology employed a segment-wise analysis, accounting for variations in breathing patterns at the night, based on wakefulness, sleep, sleep stage, and positive airway pressure device usage. This approach offers an understanding of changes in sleep status and pathophysiological states compared to measuring the AHI based on the entire night\\u0026rsquo;s sleep. Third, we assessed the performance of the Model against commonly used clinical screening questionnaires for sleep apnea, demonstrating its clinical utility. Many clinical screening tools face limitations in clinical settings owing to their low accuracy, but the Model developed in this study exhibited high accuracy and precision, potentially overcoming existing screening limitations. In conclusion, this study demonstrates that an AI model developed from EKG signals captured by a wearable Holter monitoring device offers superior screening performance for OSA compared with previous screening questionnaires. These findings suggest its potential as a practical and accessible tool, particularly for patients facing challenges during PSG.\\u003c/p\\u003e\"},{\"header\":\"Methods\",\"content\":\"\\u003cdiv id=\\\"Sec10\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eData and participants\\u003c/h2\\u003e\\u003cp\\u003eThe study participants who visited the Severance Hospital Sleep Clinic and underwent PSG between January and October 2022 were included. Inclusion criteria for adults aged 20 years and above included: 1) the presence of symptoms such as daytime sleepiness, frequent snoring, sleep apnea, fatigue, choking during sleep, frequent tossing and turning, and frequent awakening during sleep; 2) Modified Mallampatti score of grade 3 or higher, indicating difficulty during laryngotracheal intubation; and 3) presence of hypertension, heart disease, cerebrovascular disease, diabetes, or a body mass index (BMI) of 30 kg/m\\u003csup\\u003e2\\u003c/sup\\u003e or higher. Satisfying conditions 1) and 2) or conditions 1) and 3) were enough for inclusion. Pregnant women and patients with persistent atrial fibrillation, ventricular tachycardia, or implanted pacemakers were excluded. Informed consent was obtained from all eligible participants. This study was approved by the institutional review board of the Severance Hospital (No. 1-2021-0078). All the procedures were performed in accordance with the principles of the Declaration of Helsinki.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec11\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eWearable patch Holter monitoring signal\\u003c/h2\\u003e\\u003cp\\u003eThe HiCardi system, a wearable Holter EKG, facilitates extended monitoring of EKG and respiration in diverse patient groups. Comprised of the HiCardi terminal (Smart Patch), a mobile application (Smart View), and web viewer software (Live Studio), it provides a comprehensive clinical evaluation. The mobile application acts as a conduit for transmitting EKG data to a web server, and the web viewer aids the medical staff in data verification. By utilizing disposable electrodes, the device offers real-time EKG measurements accessible through both a smartphone application and a PC web viewer. To use the device, the protective film was peeled off from the disposable electrode, and the device was placed at the center of the patient's chest. The recommended attachment site is along the central sternal line, two-thirds of which are below the starting point. This device was affixed to the patients participating in the study concurrently with PSG, and the data were recorded (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003eA).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec12\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003ePolysomnography\\u003c/h2\\u003e\\u003cp\\u003eWe conducted overnight in-laboratory PSG recordings using Natus SleepWorks Software (Natus Medical Inc., USA). Patients were instructed to sleep in a controlled environment with dim lighting, temperature regulation, and noise control. The PSG recordings included electroencephalography with frontal, central, and occipital electrodes; 1-lead EKG; electromyography of extraocular eye movement, chin, and bilateral anterior tibialis muscles; nasal airflow and thermistor; peripheral oxygen saturation; sleep position; and chest and abdominal plethysmography. Sleep staging and scoring of respiratory events and movements followed the guidelines of the American Academy of Sleep Medicine (version 3.0) and were conducted by two sleep technicians with over 10 years of experience. Two neurologists (SHL and KMK), experienced in PSG interpretation, meticulously reviewed the PSG data. They reevaluated sleep staging and respiratory events, leading to the confirmation and approval of the final diagnosis and classification of OSA.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec13\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eSleep questionnaires\\u003c/h2\\u003e\\u003cp\\u003eQuestionnaires, including the Pittsburgh Sleep Quality Index (PSQI), Beck Depression Inventory-II (BDI-II), Insomnia Severity Index (ISI), Epworth Sleepiness Scale (ESS), STOP-Bang score, and Berlin questionnaire, were administered to assess the sleep habits and related problems of the participants before the PSG. The PSQI assesses sleep quality and disturbances on a scale ranging from 0\\u0026ndash;21, considering aspects such as sleep duration, efficiency, interruptions, and daytime dysfunction over a 1-month period\\u003csup\\u003e\\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e25\\u003c/span\\u003e\\u003c/sup\\u003e. BDI-II is a 21-item self-report questionnaire that assesses the severity of depressive symptoms. Respondents rated their statements based on their experiences over the past 2 weeks. The total score ranged from 0\\u0026ndash;63, with higher scores indicating more severe depressive symptoms\\u003csup\\u003e\\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e26\\u003c/span\\u003e\\u003c/sup\\u003e. The ISI is a seven-item questionnaire measuring insomnia severity over the past 2 weeks, with scores ranging from 0\\u0026ndash;28. It assesses various aspects of sleep disruption and its impact on daily life\\u003csup\\u003e\\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e27\\u003c/span\\u003e\\u003c/sup\\u003e. The ESS is an eight-item questionnaire evaluating daytime sleepiness. Participants rated their likelihood of dozing off in different situations, with scores ranging from 0\\u0026ndash;24\\u003csup\\u003e28\\u003c/sup\\u003e. The STOP-Bang questionnaire scores individuals on eight factors ranging from 0\\u0026ndash;8, including snoring, tiredness, observed apnea, high blood pressure, BMI, age (\\u0026ge;\\u0026thinsp;50 years), neck circumference, and male sex\\u003csup\\u003e\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR29\\\" class=\\\"CitationRef\\\"\\u003e29\\u003c/span\\u003e\\u003c/sup\\u003e. Higher scores indicate a higher risk of OSA. The Berlin questionnaire evaluates the risk of sleep apnea through snoring, daytime fatigue, and hypertension/obesity. It categorizes individuals as high- or low- risk\\u003csup\\u003e\\u003cspan citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e30\\u003c/span\\u003e\\u003c/sup\\u003e.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec14\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eData Pre-processing\\u003c/h2\\u003e\\u003cp\\u003eThe EKG signal collected from the wearable patch Holter device may experience some packet loss as it is transmitted over the wireless network. Therefore, the signal is compensated for by filling in the intervals where packet loss occurs with zeros so that the signal is continuous. The corrected signal uses an adaptive notch filter to remove the 60Hz band to remove powerline noise, and a high pass filter of 0.5Hz and a low pass filter of 40Hz for smooth analysis. An additional adaptive filter is applied to reduce baseline wander.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec15\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eDeep learning\\u003c/h2\\u003e\\u003cp\\u003eWe developed a 1-dimensional dilated convolutional neural network utilizing a 3-minute EKG signal as an input to predict sleep apnea (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003eB). All recordings were divided into 3-minute segments. The training and test sets were assigned by randomly shuffling all segments in a 7:3 ratio. Our Model exclusively uses raw EKG signals that have undergone z-score normalization without relying on features derived from EKG or the demographic data of the subjects.\\u003c/p\\u003e\\u003cp\\u003eThe model comprised of four convolutional blocks and an inference block. The convolutional blocks were responsible for identifying and compressing features from the EKG signals for sleep apnea detection, whereas the inference block predicted the presence or absence of sleep apnea based on the extracted features. Each convolutional block consisted of a dilated convolutional layer and a max-pooling layer, whereas the inference block included a Global Average Pooling (GAP) layer and a dense layer. Thus, the Model used comprised of ten layers. Considering the specifics of EKG signals, the Model is based on 1-D convolutional layers, but utilizes dilated convolutional layers instead of standard convolutional layers\\u003csup\\u003e\\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e31\\u003c/span\\u003e\\u003c/sup\\u003e. Dilated convolution accommodates a broader receptive field by adjusting the spacing between the filters, which is a feature controlled by the dilation rate. A dilation rate of 1 equates to standard convolution, whereas a rate of 2 introduces zero-filled spaces between filter elements, allowing broader coverage of input values with the same filter size and a deeper understanding of the input data. Due to these characteristics, each filter in the model extracts features that are effective in distinguishing between normal and apnea/hypopnea. We experimented with various dilation rates from 1 to 6 and ultimately set the dilation rate to 2. A small dilation rate captured a wider area and showed improved results; whereas, a large dilation rate was expected to fail in capturing important features owing to the spacing between filters. Each dilated convolutional layer employed a filter size of 10, with a total of 200 filters and a stride size of 1. Padding was performed to ensure equal input and output sizes of the layers. The feature maps generated from the dilated convolutional layer were processed through a rectified linear unit activation function before being input into the max-pooling layer. The pool size in the maximum pooling layer is 2, halving the size of the input feature map. The principal features selected by the max-pooling layer are then transferred to the next convolutional block.\\u003c/p\\u003e\\u003cp\\u003eThe features passing through the four convolutional blocks were fed into the GAP layer of the inference block, converting them into a 1-dimensional array corresponding to the size of the feature map extracted from the convolutional blocks. Each element of the array represents the average value at that location on the feature map. GAP reduces the dimensions by averaging the feature map, significantly lowering the Model\\u0026rsquo;s parameter count compared to a flattened layer, thereby reducing the overfitting risk. Because GAP averages the entire feature map, it preserves information irrespective of the specific locations of the main features in each map. Finally, a 1-dimensional array of features processed through GAP was utilized to output prediction values via a dense layer.\\u003c/p\\u003e\\u003cp\\u003eThe batch size is set to 64 to maximize the memory of the A100 GPU during the model training process. One of the hyper-parameters, epoch, is initially set to 100. Epoch is set to automatically terminate training without further epoch progression if the model is suspected to be overfitting by the early stopping method during the training process. As a result, training was terminated at 94 epochs by the early stopping method. The supplementary information figure \\u003cspan refid=\\\"MOESM1\\\" class=\\\"InternalRef\\\"\\u003eS1\\u003c/span\\u003e shows the respective loss and accuracy values for train and validation for each epoch. Also, some of the key hyper-parameters, optimizer and learning rate, are set to 'Nadam' optimizer and 0.001 respectively. This was chosen after experimenting with several optimizers and learning rate ratios, and the supplementary information table \\u003cspan refid=\\\"MOESM1\\\" class=\\\"InternalRef\\\"\\u003eS1\\u003c/span\\u003e shows the experimental results for each optimizer and learning rate. Table \\u003cspan refid=\\\"MOESM1\\\" class=\\\"InternalRef\\\"\\u003eS1\\u003c/span\\u003e shows training loss, training accuracy, and validation accuracy respectively, where epoch in the table refers to the point at which training was terminated by the early stopping method. We stop learning when the validation accuracy reaches a peak and the validation accuracy does not increase after about 10 epochs. In the optimizer comparison experiment, we fixed the learning rate to 0.001, and in the learning rate comparison experiment, we fixed Nadam, the best performer in the optimizer comparison experiment, as the optimizer.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec16\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eAssessment and comparison of the Model, questionnaires, and PSG classification\\u003c/h2\\u003e\\u003cp\\u003eThe segment index was defined as the ratio of the segments predicted by the final Model to be apnea segments to the total number of segments for each participant. We used Pearson\\u0026rsquo;s correlation analysis to evaluate the correlation between the Model's segment index and the PSG-derived apnea-hypopnea index (AHI)\\u003csup\\u003e\\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e32\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR33\\\" class=\\\"CitationRef\\\"\\u003e33\\u003c/span\\u003e\\u003c/sup\\u003e. The AHI predicted by the Model was obtained by applying the segment index to Pearson's correlation analysis. The variability between the Model-predicted AHI and the AHI derived from PSG was illustrated using Bland\\u0026ndash;Altman plots. In addition, for moderate or higher OSA prediction, cutoff values for sleep-related questionnaires were determined using Youden's index, calculated as sensitivity plus specificity minus one, with a range from 0 to 1\\u003csup\\u003e34\\u003c/sup\\u003e. Subsequently, we compared the Model's performance using sleep-related questionnaires based on the accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUROC) (95% confidence interval [CI]). The accuracy of the Model in classifying OSA was demonstrated using a confusion matrix with PSG, and the Cohen\\u0026rsquo;s kappa coefficient was used to evaluate the agreement between PSG and AI models in classifying OSA\\u003csup\\u003e\\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e16\\u003c/span\\u003e,\\u003cspan citationid=\\\"CR35\\\" class=\\\"CitationRef\\\"\\u003e35\\u003c/span\\u003e\\u003c/sup\\u003e. All statistical analyses were performed using Python version 3.12.1. Statistical significance was set at a two-tailed p\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.050.\\u003c/p\\u003e\\u003c/div\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003ch2\\u003eConflict of Interest statement\\u003c/h2\\u003e\\u003cp\\u003eAll authors have no conflicts to disclose.\\u003c/p\\u003e\\u003c/p\\u003e\\u003ch2\\u003eFunding\\u003c/h2\\u003e\\u003cp\\u003eThis research was supported by a grant of the Korea Health Technology R\\u0026amp;D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health \\u0026amp; Welfare, Republic of Korea (grant number: HI20C2125).\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eSHL and MK: Acquisition of data, data analysis and interpretation, and manuscript writing. SPC, YK, MKC, W-JK, and KH: Revision of manuscript critically for important intellectual content. KMK: Conception and design, data analysis and interpretation, supervision, and final approval of the version to be submitted.\\u003c/p\\u003e\\u003ch2\\u003eAcknowledgements\\u003c/h2\\u003e\\u003cp\\u003eNone\\u003c/p\\u003e\\u003ch2\\u003eData Availability\\u003c/h2\\u003e\\u003cp\\u003eAnonymized data relevant to this study will be shared upon request with a qualified investigator pending appropriate Institutional Review Board approvals.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003eKapur, V. K. \\u0026amp; Weaver, E. M. Filling in the pieces of the sleep apnea-hypertension puzzle. \\u003cem\\u003eJama\\u003c/em\\u003e \\u003cb\\u003e307\\u003c/b\\u003e, 2197\\u0026ndash;2198. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1001/jama.2012.5039\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1001/jama.2012.5039\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2012).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eGami, A. S. et al. Obstructive sleep apnea and the risk of sudden cardiac death: a longitudinal study of 10,701 adults. \\u003cem\\u003eJ. Am. Coll. Cardiol.\\u003c/em\\u003e \\u003cb\\u003e62\\u003c/b\\u003e, 610\\u0026ndash;616. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1016/j.jacc.2013.04.080\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1016/j.jacc.2013.04.080\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2013).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eDurgan, D. J. \\u0026amp; Bryan, R. M. Jr. Cerebrovascular consequences of obstructive sleep apnea. \\u003cem\\u003eJ. Am. Heart Assoc.\\u003c/em\\u003e \\u003cb\\u003e1\\u003c/b\\u003e, e000091. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1161/jaha.111.000091\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1161/jaha.111.000091\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2012).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eReichmuth, K. J., Austin, D., Skatrud, J. B. \\u0026amp; Young, T. Association of sleep apnea and type II diabetes: a population-based study. \\u003cem\\u003eAm. J. Respir Crit. Care Med.\\u003c/em\\u003e \\u003cb\\u003e172\\u003c/b\\u003e, 1590\\u0026ndash;1595. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1164/rccm.200504-637OC\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1164/rccm.200504-637OC\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2005).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eEngleman, H. M., Douglas, N. J. \\u0026amp; Sleep 4: Sleepiness, cognitive function, and quality of life in obstructive sleep apnoea/hypopnoea syndrome. \\u003cem\\u003eThorax\\u003c/em\\u003e 59, 618\\u0026ndash;622 (2004). \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1136/thx.2003.015867\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1136/thx.2003.015867\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBenjafield, A. V. et al. Estimation of the global prevalence and burden of obstructive sleep apnoea: a literature-based analysis. \\u003cem\\u003eLancet Respir Med.\\u003c/em\\u003e \\u003cb\\u003e7\\u003c/b\\u003e, 687\\u0026ndash;698. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1016/s2213-2600(19)30198-5\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1016/s2213-2600(19)30198-5\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2019).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eZou, J. et al. An effective model for screening obstructive sleep apnea: a large-scale diagnostic study. \\u003cem\\u003ePLoS One\\u003c/em\\u003e. \\u003cb\\u003e8\\u003c/b\\u003e, e80704. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1371/journal.pone.0080704\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1371/journal.pone.0080704\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2013).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eAntic, N. A. et al. A randomized controlled trial of nurse-led care for symptomatic moderate-severe obstructive sleep apnea. \\u003cem\\u003eAm. J. Respir Crit. Care Med.\\u003c/em\\u003e \\u003cb\\u003e179\\u003c/b\\u003e, 501\\u0026ndash;508. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1164/rccm.200810-1558OC\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1164/rccm.200810-1558OC\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2009).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eB\\u0026ouml;hning, N. et al. Comparability of pulse oximeters used in sleep medicine for the screening of OSA. \\u003cem\\u003ePhysiol. Meas.\\u003c/em\\u003e \\u003cb\\u003e31\\u003c/b\\u003e, 875\\u0026ndash;888. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1088/0967-3334/31/7/001\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1088/0967-3334/31/7/001\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2010).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eDawson, A. et al. Type III home sleep testing versus pulse oximetry: is the respiratory disturbance index better than the oxygen desaturation index to predict the apnoea-hypopnoea index measured during laboratory polysomnography? \\u003cem\\u003eBMJ Open.\\u003c/em\\u003e \\u003cb\\u003e5\\u003c/b\\u003e, e007956. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1136/bmjopen-2015-007956\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1136/bmjopen-2015-007956\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2015).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBazoukis, G. et al. Application of artificial intelligence in the diagnosis of sleep apnea. \\u003cem\\u003eJ. Clin. Sleep. Med.\\u003c/em\\u003e \\u003cb\\u003e19\\u003c/b\\u003e, 1337\\u0026ndash;1363. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.5664/jcsm.10532\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.5664/jcsm.10532\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2023).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eMurali, S. et al. Cardiac Ambulatory Monitoring: New Wireless Device Validated Against Conventional Holter Monitoring in a Case Series. \\u003cem\\u003eFront. Cardiovasc. Med.\\u003c/em\\u003e \\u003cb\\u003e7\\u003c/b\\u003e, 587945. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.3389/fcvm.2020.587945\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.3389/fcvm.2020.587945\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2020).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eDuncker, D. et al. Smart Wearables for Cardiac Monitoring-Real-World Use beyond Atrial Fibrillation. \\u003cem\\u003eSens. (Basel)\\u003c/em\\u003e. 21. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.3390/s21072539\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.3390/s21072539\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2021).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eNagappa, M. 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. \\u003cem\\u003ePLoS One\\u003c/em\\u003e. \\u003cb\\u003e10\\u003c/b\\u003e, e0143697. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1371/journal.pone.0143697\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1371/journal.pone.0143697\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2015).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eJeon, H. J., Bang, Y. R. \\u0026amp; Yoon, I. Y. A validation study on three screening questionnaires for obstructive sleep apnea in a Korean community sample. \\u003cem\\u003eSleep. Breath.\\u003c/em\\u003e \\u003cb\\u003e23\\u003c/b\\u003e, 969\\u0026ndash;977. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1007/s11325-018-1748-x\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1007/s11325-018-1748-x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2019).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eCohen, J. A coefficient of agreement for nominal scales. \\u003cem\\u003eEduc. Psychol. Meas.\\u003c/em\\u003e \\u003cb\\u003e20\\u003c/b\\u003e, 37\\u0026ndash;46 (1960).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eSalari, N. et al. Detection of sleep apnea using Machine learning algorithms based on ECG Signals: A comprehensive systematic review. \\u003cem\\u003eExpert Syst. Appl.\\u003c/em\\u003e \\u003cb\\u003e187\\u003c/b\\u003e, 115950 (2022).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eErdenebayar, U., Kim, Y. J., Park, J. U., Joo, E. Y. \\u0026amp; Lee, K. J. Deep learning approaches for automatic detection of sleep apnea events from an electrocardiogram. \\u003cem\\u003eComput. Methods Programs Biomed.\\u003c/em\\u003e \\u003cb\\u003e180\\u003c/b\\u003e, 105001. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1016/j.cmpb.2019.105001\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1016/j.cmpb.2019.105001\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2019).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eHassan, A. R. \\u0026amp; Haque, M. A. Computer-aided obstructive sleep apnea screening from single-lead electrocardiogram using statistical and spectral features and bootstrap aggregating. \\u003cem\\u003eBiocybernetics Biomedical Eng.\\u003c/em\\u003e \\u003cb\\u003e36\\u003c/b\\u003e, 256\\u0026ndash;266 (2016).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBabaeizadeh, S., White, D. P., Pittman, S. D. \\u0026amp; Zhou, S. H. Automatic detection and quantification of sleep apnea using heart rate variability. \\u003cem\\u003eJ. Electrocardiol.\\u003c/em\\u003e \\u003cb\\u003e43\\u003c/b\\u003e, 535\\u0026ndash;541. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1016/j.jelectrocard.2010.07.003\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1016/j.jelectrocard.2010.07.003\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2010).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eSingstad, B. J. \\u0026amp; Tronstad, C. in \\u003cem\\u003e2020 computing in cardiology.\\u003c/em\\u003e 1\\u0026ndash;4 (IEEE).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eChang, H. Y., Yeh, C. Y., Lee, C. T. \\u0026amp; Lin, C. C. A Sleep Apnea Detection System Based on a One-Dimensional Deep Convolution Neural Network Model Using Single-Lead Electrocardiogram. \\u003cem\\u003eSens. (Basel)\\u003c/em\\u003e. 20. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.3390/s20154157\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.3390/s20154157\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2020).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eChung, F., Abdullah, H. R., Liao, P. \\u0026amp; STOP-Bang Questionnaire A Practical Approach to Screen for Obstructive Sleep Apnea. \\u003cem\\u003eChest\\u003c/em\\u003e \\u003cb\\u003e149\\u003c/b\\u003e, 631\\u0026ndash;638. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1378/chest.15-0903\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1378/chest.15-0903\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2016).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eTan, A. et al. Predicting obstructive sleep apnea using the STOP-Bang questionnaire in the general population. \\u003cem\\u003eSleep. Med.\\u003c/em\\u003e \\u003cb\\u003e27\\u0026ndash;28\\u003c/b\\u003e, 66\\u0026ndash;71. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1016/j.sleep.2016.06.034\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1016/j.sleep.2016.06.034\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2016).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBuysse, D. J., Reynolds, C. F. 3, Monk, T. H., Berman, S. R., Kupfer, D. J. \\u0026amp; rd, \\u0026amp; The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. \\u003cem\\u003ePsychiatry Res.\\u003c/em\\u003e \\u003cb\\u003e28\\u003c/b\\u003e, 193\\u0026ndash;213. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1016/0165-1781(89)90047-4\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1016/0165-1781(89)90047-4\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (1989).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBeck, A. T., Steer, R. A., Ball, R. \\u0026amp; Ranieri, W. Comparison of Beck Depression Inventories -IA and -II in psychiatric outpatients. \\u003cem\\u003eJ. Pers. Assess.\\u003c/em\\u003e \\u003cb\\u003e67\\u003c/b\\u003e, 588\\u0026ndash;597. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1207/s15327752jpa6703_13\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1207/s15327752jpa6703_13\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (1996).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eMorin, C. M., Belleville, G., B\\u0026eacute;langer, L. \\u0026amp; Ivers, H. The Insomnia Severity Index: psychometric indicators to detect insomnia cases and evaluate treatment response. \\u003cem\\u003eSleep\\u003c/em\\u003e \\u003cb\\u003e34\\u003c/b\\u003e, 601\\u0026ndash;608. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1093/sleep/34.5.601\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1093/sleep/34.5.601\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2011).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eDoneh, B. Epworth Sleepiness Scale. \\u003cem\\u003eOccup. Med. (Lond)\\u003c/em\\u003e. \\u003cb\\u003e65\\u003c/b\\u003e, 508. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1093/occmed/kqv042\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1093/occmed/kqv042\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2015).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eHwang, M. et al. Validation of the STOP-Bang questionnaire as a preoperative screening tool for obstructive sleep apnea: a systematic review and meta-analysis. \\u003cem\\u003eBMC Anesthesiol\\u003c/em\\u003e. \\u003cb\\u003e22\\u003c/b\\u003e, 366. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1186/s12871-022-01912-1\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1186/s12871-022-01912-1\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (2022).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eNetzer, N. C., Stoohs, R. A., Netzer, C. M., Clark, K. \\u0026amp; Strohl, K. P. Using the Berlin Questionnaire to identify patients at risk for the sleep apnea syndrome. \\u003cem\\u003eAnn. Intern. Med.\\u003c/em\\u003e \\u003cb\\u003e131\\u003c/b\\u003e, 485\\u0026ndash;491. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.7326/0003-4819-131-7-199910050-00002\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.7326/0003-4819-131-7-199910050-00002\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (1999).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eYu, F. \\u0026amp; Koltun, V. Multi-scale context aggregation by dilated convolutions. \\u003cem\\u003earXiv preprint arXiv:1511.07122\\u003c/em\\u003e (2015).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eStigler, S. M. Francis Galton's account of the invention of correlation. \\u003cem\\u003eStatistical Science\\u003c/em\\u003e, 73\\u0026ndash;79 (1989).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eBland, J. M. \\u0026amp; Altman, D. G. Statistical methods for assessing agreement between two methods of clinical measurement. \\u003cem\\u003eLancet\\u003c/em\\u003e \\u003cb\\u003e1\\u003c/b\\u003e, 307\\u0026ndash;310 (1986).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eYouden, W. J. Index for rating diagnostic tests. \\u003cem\\u003eCancer\\u003c/em\\u003e \\u003cb\\u003e3\\u003c/b\\u003e, 32\\u0026ndash;35. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org:10.1002/1097-0142(1950)3:1\\u0026lt;32::aid-cncr2820030106\\u0026gt;3.0.co;2-3\\u003c/span\\u003e\\u003cspan address=\\\"https://doi.org:10.1002/1097-0142(1950)3:1%3C32::aid-cncr2820030106%3E3.0.co;2-3\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e (1950).\\u003c/span\\u003e\\u003c/li\\u003e\\u003cli\\u003e\\u003cspan\\u003eStehman, S. V. Selecting and interpreting measures of thematic classification accuracy. \\u003cem\\u003eRemote Sens. Environ.\\u003c/em\\u003e \\u003cb\\u003e62\\u003c/b\\u003e, 77\\u0026ndash;89 (1997).\\u003c/span\\u003e\\u003c/li\\u003e\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":false,\"hideJournal\":false,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"scientific-reports\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"scirep\",\"sideBox\":\"Learn more about [Scientific Reports](http://www.nature.com/srep/)\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Scientific Reports\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Scientific Reports\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true},\"keywords\":\"Obstructive sleep apnea, Electrocardiogram, Deep learning, Wearable devices, Convolutional neural network, Polysomnography validation\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-7537553/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-7537553/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eObstructive sleep apnea (OSA) is a highly prevalent disorder that remains underdiagnosed due to the limited accessibility and high cost of polysomnography (PSG), the current diagnostic standard. This study presents a deep learning model that detects OSA from single-lead electrocardiogram (ECG) signals acquired via a wearable patch Holter device. We collected ECG data from 92 adult patients undergoing overnight PSG at a sleep clinic. A 1-dimensional dilated convolutional neural network (1D-CNN) was trained using 3-minute ECG segments and validated against PSG-derived apnea-hypopnea index (AHI) scores. The model achieved an accuracy of 81.1%, precision of 81.9%, recall of 82.5%, F1-score of 82.2%, and an area under the receiver operating characteristic curve (AUROC) of 0.875. The predicted AHI was strongly correlated with PSG AHI (r\\u0026thinsp;=\\u0026thinsp;0.847), and the model outperformed conventional screening questionnaires such as STOP-Bang in identifying moderate-to-severe OSA (AUROC\\u0026thinsp;=\\u0026thinsp;0.888). These results demonstrate the feasibility of using wearable ECG and deep learning for accurate and scalable OSA detection. This approach may offer a non-invasive and cost-effective alternative to PSG in both clinical and community-based screening settings.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Sleep Apnea Detection Using Wearable ECG and Deep Learning: Validation with Polysomnography\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-09-17 18:42:08\",\"doi\":\"10.21203/rs.3.rs-7537553/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0},{\"type\":\"decision\",\"content\":\"Revision requested\",\"date\":\"2025-10-20T06:21:06+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-10-19T04:03:05+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-10-17T09:08:15+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-10-14T05:20:01+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"120960199890520690468887561736643633517\",\"date\":\"2025-10-06T08:10:31+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"329989113402332657015747890035208237866\",\"date\":\"2025-10-03T08:23:08+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"155735558207662322694354896239252646622\",\"date\":\"2025-10-03T02:37:06+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"253654626779903234475164333700725122695\",\"date\":\"2025-10-02T02:04:41+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"238298805293552552860120052060972489228\",\"date\":\"2025-10-01T16:50:33+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"17520677204339636253135369262961046651\",\"date\":\"2025-10-01T13:58:45+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewersInvited\",\"content\":\"\",\"date\":\"2025-09-09T15:18:02+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvited\",\"content\":\"\",\"date\":\"2025-09-09T11:41:25+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorAssigned\",\"content\":\"\",\"date\":\"2025-09-08T06:03:30+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"checksComplete\",\"content\":\"\",\"date\":\"2025-09-06T04:47:03+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"submitted\",\"content\":\"Scientific Reports\",\"date\":\"2025-09-04T15:10:03+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"scientific-reports\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"scirep\",\"sideBox\":\"Learn more about [Scientific Reports](http://www.nature.com/srep/)\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Scientific Reports\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Scientific Reports\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"ba78f90e-e04d-4f55-bb8b-922b2c94039b\",\"owner\":[],\"postedDate\":\"September 17th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"in-revision\",\"subjectAreas\":[{\"id\":54687972,\"name\":\"Health sciences/Cardiology\"},{\"id\":54687973,\"name\":\"Biological sciences/Computational biology and bioinformatics\"},{\"id\":54687974,\"name\":\"Health sciences/Diseases\"},{\"id\":54687975,\"name\":\"Health sciences/Health care\"},{\"id\":54687976,\"name\":\"Health sciences/Medical research\"}],\"tags\":[],\"updatedAt\":\"2026-05-13T08:43:17+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-09-17 18:42:08\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-7537553\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-7537553\",\"identity\":\"rs-7537553\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}