Polysomnography (PSG) is the gold standard for diagnosing obstructive sleep apnea (OSA). However, it is limited by its reliance on sleep state, costly equipment, and complex operational requirements. Existing deep learning approaches for OSA detection often utilize limited modalities such as electrocardiograms and snoring audio, frequently overlooking the potential of facial image data and other multidimensional inputs. To address these limitations, we propose a data-driven diagnostic framework, MIMAR-OSA, which leverages three distinct modalities: basic physiological measurements, facial image data, and snoring audio collected by patients using portable devices. Our framework employs attention grids in the visual domain to identify facial key points from images at various angles, a graph attention network to emphasize and connect critical facial features across images, and semantic processing of patient’s basic physical signs information in the linguistic domain. For the audio domain, we utilize diffWave to reconstruct missing audio components. Comprehensive data integration and analysis are facilitated by a fusion transformer combined with a mixture of expert (MoE) models. Our framework was validated on a university hospital OSA dataset, achieving a 92.24% accuracy. Additionally, testing our MIMAR-OSA model on open-source multimodal learning benchmarks demonstrated its superiority in multi-classification tasks over other state-of-the-art models, confirming its broad applicability and extensibility across various medical settings.

BackgroundAccurate prediction of therapeutic pressure for Continuous Positive Airway Pressure (CPAP) therapy is essential for effective treatment of Obstructive Sleep Apnea (OSA). Existing methods often rely on complex sleep-related parameters and small sample sizes, limiting their generalizability. This study aims to develop a more accessible, data-driven model using readily available demographic and physiological variables to predict CPAP pressure, improving both accuracy and scalability.MethodsWe employed a machine learning approach, integrating decision trees, gradient boosting algorithms (LightGBM, XGBoost, CatBoost), and neural networks to predict therapeutic pressure. Forward selection based on the Akaike Information Criterion (AIC) was used to identify the most relevant variables. The model was trained on a dataset of 2,092 patients, with model performance assessed using mean absolute error (MAE).ResultsThe most influential variables identified were BMI, neck circumference, and waist-to-hip ratio. Among the algorithms, LightGBM achieved the highest predictive accuracy, with the lowest MAE. Ensemble methods, such as voting, did not improve performance beyond LightGBM alone. Subsample analyses revealed that prediction accuracy varied across BMI ranges and ventilator brands.ConclusionsThe study demonstrates that BMI and other physical parameters play a pivotal role in determining CPAP pressure, offering a simplified yet effective prediction model. This approach has significant potential for clinical applications, particularly in resource-limited settings, where access to complex sleep studies may be restricted. Future research could enhance the model by incorporating real-time physiological data and expanding data collection to diverse populations.

Effective pre-polysomnography screening for obstructive sleep apnea-hypopnea syndrome (OSAHS) requires combining clinical risk factors with visible craniofacial and neck cues. Directly prompting general-purpose multimodal foundation models for medical yes/no decisions can yield unstable, poorly calibrated outputs. We propose EviOSAHS, an evidence-grounded multimodal reasoning framework that separates image-only anatomical evidence acquisition from final clinical adjudication. Each frontal facial image is decomposed into seven fixed anatomical queries covering the neck, chin, mouth, face/neck fat, lower jaw, midface, and nose. Visual responses are converted into structured evidence cards recording target anatomy, visibility, risk direction, evidence strength, confidence, and a concise summary. These cards are combined with a cleaned clinical profile only in the final stage, where a large language model performs balanced binary screening adjudication. We evaluated EviOSAHS on a 642-subject cohort, mapping normal subjects to screening-negative and mild, moderate, or severe OSAHS subjects to screening-positive. EviOSAHS achieved 88.47

Background:Sleep apnea-hypopnea syndrome (SAHS) has high prevalence and cardiovascular/cerebrovascular comorbidities. Polysomnography (PSG) is the diagnostic gold standard but limited by complexity, cost, and accessibility; smart wearables are convenient but less accurate. This study validated an AI-based snoring-sound model and compared its diagnostic performance with a wearable, using PSG as reference. Methods:Adults with suspected SAHS (n = 134) were included and underwent overnight PSG, with simultaneous snoring sound and wearable signal recording. The snoring algorithm used short-time Fourier transform (STFT) spectrograms and a multiscale encoder-attention-decoder neural network for classification. Model performance was assessed using correlation analysis, intraclass correlation coefficient (ICC), Bland-Altman plots, and receiver operating characteristic (ROC) analyses. Accuracy, sensitivity, specificity, and area under the curve (AUC) values were computed for both models. Results:Disease severity was associated with variation in BMI (p < 0.001) and decreasing oxygen saturation (p < 0.001). The snoring model showed stronger correlation and higher agreement with PSG-derived AHI (r = 0.79, ICC = 0.753) than the wearable (r = 0.68). Although the snoring model yielded consistently higher AUCs, particularly in mild SAHS (AUC = 0.83 vs. 0.72, p = 0.07), the differences between methods were not statistically significant (all p > 0.05). Both models achieved excellent discriminative performance in severe SAHS (AUC ≥ 0.90). Conclusion:The AI-driven snoring-sound model demonstrated comparable overall performance and numerically better performance in certain settings, particularly in identifying mild SAHS.

INTRODUCTION:The pharmacological profile of remimazolam makes it particularly suitable for patients living with obesity. However, how its pharmacokinetics are influenced by varying degrees of obesity and obstructive sleep apnoea severity remains unclear. This study aimed to characterise remimazolam pharmacokinetics in volunteers living with different levels of obesity and obstructive sleep apnoea. METHODS:Stepwise escalating infusion of remimazolam was administered until five consecutive modified observer's assessment of alertness and sedation scores of 1 were recorded. Arterial blood samples were collected before infusion, at each infusion rate adjustment step, 5 minutes after each adjustment and several time points post-infusion discontinuation. Drug concentrations were quantified using ultra-high-performance liquid chromatography coupled with triple quadrupole tandem mass spectrometry. A joint population pharmacokinetic model characterising remimazolam and its metabolite CNS7054 was developed. RESULTS:Forty-four adult volunteers, 30 (68.2%) with obstructive sleep apnoea, were studied. Remimazolam pharmacokinetics were best described by a three-compartment model; a one-compartment model with transit compartment described CNS7054. External validation confirmed robust prediction performance. Clearance of remimazolam was high (1.09 l.min-1, relative standard error = 4%). Total body weight significantly affected clearance and metabolite volume of distribution. No other covariates, including obstructive sleep apnoea severity, had significant impact. In patients living with class 2/3 obesity, context-sensitive half-time after a 4-h infusion was short (4.02 min). Dosing based on lean/ideal body weight for induction, or adjusted body weight for maintenance, in patients living with class 2/3 obesity yielded pharmacokinetic profiles comparable with those achieved with total body weight dosing in normal weight patients. DISCUSSION:Body size significantly alters remimazolam pharmacokinetics and individualised weight-based dosing is essential for populations with obesity. Obstructive sleep apnoea severity does not affect drug disposition; however, its pharmacodynamic effects require further investigation.

Obstructive sleep apnea-hypopnea syndrome (OS-AHS) is a common sleep disorder caused by upper airway blockage, leading to oxygen deprivation and disrupted sleep. Traditional diagnosis using polysomnography (PSG) is expensive, time-consuming, and uncomfortable. Existing deep learning methods using facial image analysis lack accuracy due to poor facial feature capture and limited sample sizes. To address this, we propose a multimodal dual encoder model that integrates visual and language inputs for automated OSAHS diagnosis. The model balances data using randomOverSampler (ROS), extracts key facial features with attention grids, and converts basic physiological data into meaningful text. Cross attention combines image and text data for better feature extraction, and ordered regression loss ensures stable learning. Our approach improves diagnostic efficiency and accuracy, achieving 91.3% top-1 accuracy in a four class severity classification task, demonstrating state of the art performance. Code is available at https://github.com/luboyan6/VTA-OSAHS.

Polysomnography (PSG) is the gold standard for diagnosing obstructive sleep apnea-hypopnea syndrome (OSAHS), but it relies heavily on manual scoring by experienced clinicians, which is time-consuming and costly. This study presents an automated method for detecting sleep apnea and hypopnea events using overnight airflow and blood oxygen saturation signals. Apnea episodes are identified through local range (LR) analysis of airflow, while rapid oxygen desaturation supports hypopnea detection. LR thresholds were trained on data from 100 patients and validated on an independent cohort of 43 patients. The system's performance was compared to clinician annotations using the apnea-hypopnea index (AHI). The results demonstrated high consistency with manual scoring, highlighting the method's potential for home-based monitoring and large-scale screening. This approach offers a lightweight and scalable alternative to traditional diagnostic procedures, reducing dependence on complex equipment and human interpretation.

Sleep Apnea Syndrome (SAS) is a serious respiratory disorder that can lead to a range of complications, including hypertension, arrhythmias, cognitive impair- ment, and metabolic disturbances. Due to the insidious nature of its symptoms, patients often fail to recognize the condition, and clinical screening is both time- consuming and resource-intensive. To address these challenges, we have developed a comprehensive dataset that integrates data from Polysomnography (PSG) devices with synchronized audio recordings. This dataset has been rigorously annotated by expert medical professionals based on PSG monitoring data, ensur- ing its accuracy and reliability. Our objective is to provide a publicly available, standardized, and high-quality data resource for the development and applica- tion of deep learning models in the field of sleep apnea syndrome. This dataset is designed to enhance diagnostic accuracy and efficiency while promoting advanced scientific research and technological innovation in this domain.

Obstructive sleep apnea-hypopnea syndrome (OSAHS) is a prevalent chronic disorder that affects sleep quality and general health. The current diagnostic methods, primarily polysomnography (PSG), are laborious. Furthermore, audio-based methods for diagnosing OSAHS face limited sample sizes and neglect patients’ physiological signs and medical histories. To address these challenges, we introduce a data-driven framework called DFNet, which also considers patients’ medical histories and health indicators. DFNet incorporates an automated audio segmentation- and labeling-based preprocessing procedure to reduce expert annotation costs and subjective errors. We employed random convolutional kernels based on receptive fields for audio feature extraction purposes. These kernels captured both local and global features within the input audio. Additionally, for the first time, we introduced a medical language model that utilizes patients’ medical histories and physiological information as covariates to enhance features. We extensively validated DFNet on an OSAHS dataset obtained from a collaborative university hospital. Our framework classified patients into four categories according to their OSAHS severity: normal, mild, moderate, and severe. DFNet achieved state-of-the-art performance, with a four-class accuracy of 84.12 https://github.com/testlbin/DFNet .

Sleep stage prediction is a critical task in medical diagnostics, such as for sleep disorders like Obstructive Sleep Apnea-Hypopnea Syndrome (OSAHS). Traditionally, this task involves analyzing Electroencephalogram (EEG) signals and classifying the stages based on general features, often relying on medical expertise. However, this process is prone to bias and variance, as clinicians incorporate subjective experience into their predictions. In recent years, multimodal large language models (MLLMs) have demonstrated significant advancements, particularly in medical applications, outperforming traditional methods in many domains. Despite their promising potential, MLLMs are sensitive to high memorization effects and require high-quality, well-labeled data for fine-tuning. Label noise, commonly present in real-world datasets, can severely hinder their performance and robustness. Consequently, directly applying MLLMs to sleep stage prediction using noisy EEG labels presents a challenge. In this paper, we introduce a novel framework for sleep stage prediction using EEG data under label noise, leveraging the p ower of MLLMs. Our approach integrates multi-perspective agreement techniques to identify high-quality samples based on the prior knowledge embedded in MLLMs. We then employ a self-training method to enhance prediction accuracy despite the presence of label noise. We validate our framework using real patient EEG data in sleep stage prediction tasks, and the results demonstrate that our approach is both robust and accurate under label noise, outperforming other state-of-theart robust learning methods. Our code will be made publicly available at https://github.com/Leonard- zc/MICCAI2025- RSSP.

Sleep staging plays a vital role in diagnosing sleep disorders, with obstructive sleep apnea (OSA) being one of the most prevalent conditions. Clinical data from OSA patients, including sleep duration, age, gender, BMI, and sleep position, has been found to significantly correlate with sleep stages. In this study, we proposed a large language model (LLM)-based clinical information analysis method utilizing OpenAI’s latest model, ChatGPT-4o, to analyze clinical data from 2570 OSA patients rigorously selected by medical experts. The results of this method indicate that sleep position is the most influential factor affecting sleep staging accuracy, accounting for 74.8 https://github.com/yykk187/spsleepnet .

ObjectiveThis study aims to explore the effects of obstructive sleep apnea (OSA) on hemorheological parameters.MethodsBetween February 2023 and March 2024, 882 patients were screened, and 787 were included in the study. The participants were divided into a non-OSA group (72 individuals) and an OSA group (715 individuals). Propensity score overlap weighting was used to balance the clinical variables, red blood cell count, and blood biochemistry results between the two groups, and regression analysis was conducted to assess the relationship between OSA and hemorheological parameters. Sensitivity analysis was conducted using propensity score covariate adjustment to ensure the robustness of the results.ResultsCompared to non-OSA patients, OSA patients exhibited a significant increase in whole blood viscosity at high shear rates by 0.18 mPa.s (95% CI: [0.03, 0.32]) and in plasma viscosity by 0.05 mPa.s (95% CI: [0.01, 0.10]). However, there was no significant difference in whole blood viscosity at low shear rates between the two groups (95%CI: [-0.01, 0.56]).ConclusionAfter adjusting the effects of comorbidities such as hypertension, diabetes, and cardiovascular diseases, OSA still had a significant impact on high-shear whole blood viscosity and plasma viscosity.

IntroductionObstructive Sleep Apnea-Hypopnea Syndrome (OSAHS) is a common sleep-related breathing disorder that significantly impacts the daily lives of patients. Currently, the diagnosis of OSAHS relies on various physiological signal monitoring devices, requiring a comprehensive Polysomnography (PSG). However, this invasive diagnostic method faces challenges such as data fluctuation and high costs. To address these challenges, we propose a novel data-driven Audio-Semantic Multi-Modal model for OSAHS severity classification (i.e., ASMM-OSA) based on patient snoring sound characteristics.MethodsIn light of the correlation between the acoustic attributes of a patient's snoring patterns and their episodes of breathing disorders, we utilize the patient's sleep audio recordings as an initial screening modality. We analyze the audio features of snoring sounds during the night for subjects suspected of having OSAHS. Audio features were augmented via PubMedBERT to enrich their diversity and detail and subsequently classified for OSAHS severity using XGBoost based on the number of sleep apnea events.ResultsExperimental results using the OSAHS dataset from a collaborative university hospital demonstrate that our ASMM-OSA audio-semantic multimodal model achieves a diagnostic level in automatically identifying sleep apnea events and classifying the four-class severity (normal, mild, moderate, and severe) of OSAHS.DiscussionOur proposed model promises new perspectives for non-invasive OSAHS diagnosis, potentially reducing costs and enhancing patient quality of life.

Polysomnography is the diagnostic gold standard for obstructive sleep apnea-hypopnea syndrome (OSAHS), requiring medical professionals to analyze apnea-hypopnea events from multidimensional data throughout the sleep cycle. This complex process is susceptible to variability based on the clinician’s experience, leading to potential inaccuracies. Existing automatic diagnosis methods often overlook multimodal physiological signals and medical prior knowledge, leading to limited diagnostic capabilities. This study presents a novel heterogeneous graph convolutional fusion network (HeteroGCFNet) leveraging multimodal physiological signals and domain knowledge for automated OSAHS diagnosis. This framework constructs two types of graph representations: physical space graphs, which map the spatial layout of sensors on the human body, and process knowledge graphs which detail the physiological relationships among breathing patterns, oxygen saturation, and vital signals. The framework leverages heterogeneous graph convolutional neural networks to extract both localized and global features from these graphs. Additionally, a multi-head fusion module combines these features into a unified representation for effective classification, enhancing focus on relevant signal characteristics and cross-modal interactions. This study evaluated the proposed framework on a large-scale OSAHS dataset, combined from publicly available sources and data provided by a collaborative university hospital. It demonstrated superior diagnostic performance compared to conventional machine learning models and existing deep learning approaches, effectively integrating domain knowledge with data-driven learning to produce explainable representations and robust generalization capabilities, which can potentially be utilized for clinical use. Code is available at https://github.com/AmbitYuki/HeteroGCFNet .

Objectives: Polysomnography was class I test for who was suspected of obstructive sleep apnea (OSA) which would cost lots of time and money. This study aimed to develop a nomogram model mainly based on oxygen and blood routine indicators to predict OSA. Methods: We retrospectively analyzed 685 patients with suspected OSA at our hospital. Multivariate analysis was used to construct a nomogram. The performance of the nomogram was assessed using calibration and discrimination. Results: The multivariate analysis identified age, gender, body mass index, mean pulse oxygen saturation, percent nighttime with oxygen saturation less than 90%, red blood cell, hematocrit, and red blood cell distribution width SD as significant factors (P < .05). A nomogram was created for the prediction of OSA using these clinical parameters and was internally validated using a bootstrapping method. Our nomogram model showed good discrimination and calibration in terms of predicting OSA, and had a C-index of 0.935 [95% confidence interval (CI), 0.917-0.954] according to the internal validation. Discrimination and calibration in the validation group were also good (C-index, 0.957; 95% CI, 0.930-0.984). Conclusion: The newly developed nomogram can effectively help physicians make better clinical decisions, which may save a lot of time and costs.

The aim of this study was to characterise the whole-body 50 kHz phase angle in patients with obstructive sleep apnoea (OSA) and to estimate it using other nutrition-related physical examination indices. We enrolled 117 patients with OSA who underwent polysomnography. Patients underwent body composition measurements, physical examinations, and nutrition-related laboratory tests, and were grouped according to neck circumference (NC) quartiles to analyse the correlation between NC and nutritional assessment index values. A significant positive correlation was observed between NC and various nutritional indicators, including height and whole-body 50 kHz phase angle. An increase in NC resulted in a corresponding increase in weight, body mass index, waist circumference, arm muscle circumference, whole-body 50 kHz phase angle, apnoea–hypopnea index, and the longest apnoea time. Further stepwise regression and multiple linear regression analyses revealed: whole-body 50 kHz phase angle =(0.126×waist circumference + 0.345×arm muscle circumference − 1.420- NC)/(-1.496). The whole-body 50 kHz phase angle in patients with OSA can be estimated using NC, arm muscle circumference, and waist circumference, helping assess the combined nutritional, inflammatory, and immune status of patients with OSA.

The study aimed to assess the performance of the PVT in patients with suspected OSA, evaluate its role in population screening for OSA. The NoSAS, STOP-Bang, ESS scores and PVT tests were performed after suspected OSA patients’ admission, followed by PSG. Then we compared the PVT results, calculated the sensitivity, specificity and ROC curve of PVT, and analyzed the accuracy of STOP-Bang and NoSAS questionnaire combined with PVT in predicting OSA. A total of 308 patients were divided into four groups based on AHI: primary snoring (2.74 ± 1.4 events/h, n = 37); mild OSA (9.96 ± 3.25 events/h, n = 65); moderate OSA (22.41 ± 4.48 events/h, n = 76); and, severe OSA (59.42 ± 18.37 events/h, n = 130). There were significant differences in PVT lapses (p < 0.001) and reaction time (RT, p = 0.03) among the four groups. The PVT lapses and RT were positively correlated with AHI (p < 0.001) and ODI (p < 0.001), and negatively correlated with LSpO2 (p < 0.001). When diagnosing OSA (AHI ≥ 5 events/h), the AUCs of PVT, ESS, STOP-Bang, and NoSAS were 0.679, 0.579, 0.727, and 0.653, respectively; the AUCs of STOP-Bang and NoSAS combined with PVT increased. After combined PVT, the diagnostic specificity of STOP-Bang and NoSAS at nodes with AHI ≥ 5, ≥ 15 and ≥ 30 events/h increased to varying degrees. Patients with OSA exhibited impairment in the PVT, and the combination of the PVT and STOP-Bang or NoSAS scores can improve the diagnostic efficacy and specificity for OSA.

Integrating deep neural networks with the Hawkes process has significantly improved predictive capabilities in finance, health informatics, and information technology. Nevertheless, these models often face challenges in real-world settings, particularly due to substantial label noise. This issue is of significant concern in the medical field, where label noise can arise from delayed updates in electronic medical records or misdiagnoses, leading to increased prediction risks. Our research indicates that deep Hawkes process models exhibit reduced robustness when dealing with label noise, particularly when it affects both event types and timing. To address these challenges, we first investigate the influence of label noise in approximated intensity functions and present a novel framework, the Robust Deep Hawkes Process (RDHP), to overcome the impact of label noise on the intensity function of Hawkes models, considering both the events and their occurrences. We tested RDHP using multiple open-source benchmarks with synthetic noise and conducted a case study on obstructive sleep apnea-hypopnea syndrome (OSAHS) in a real-world setting with inherent label noise. The results demonstrate that RDHP can effectively perform classification and regression tasks, even in the presence of noise related to events and their timing. To the best of our knowledge, this is the first study to successfully address both event and time label noise in deep Hawkes process models, offering a promising solution for medical applications, specifically in diagnosing OSAHS.

This study investigated the roles of P-selectin and Clara cell secretory protein 16 (CC16) levels in the pathogenesis of severe adenovirus (ADV) pneumonia in children and evaluated their ability to predict disease. Fifty-one children (age, 1-5 years) with ADV pneumonia who were admitted to Xiamen Children's Hospital were included in this study and divided into the mild group (24 patients) and severe group (27 patients). A control group comprising healthy children of the same age who underwent routine physical examinations during the same period (30 patients) was also included. The univariate analysis demonstrated that the levels of the white blood cell count and C-reactive protein, procalcitonin, d-dimer, and P-selectin were increased in a severe group compared with a mild group, while CC16 levels were significantly decreased (p < 0.05). The logistic regression analysis revealed that P-selectin and CC16 levels were independent risk factors for severe ADV pneumonia in children. The areas under the ROC curves suggested that P-selectin and CC16 exhibited high predictive value for severe ADV pneumonia. P-selectin values more than 898.58 pg/mL and CC16 values less than 11.355 ng/mL predicted severe ADV pneumonia. P-selectin and CC16 levels are correlated with the severity of ADV pneumonia in children.