A survey on pre-training requirements for deep learning models to detect obstructive sleep apnea events

Over the last two decades, social media platforms have grown dramatically. Twitter and Facebook are the two most popular social media platforms, with millions of active users posting billions of messages daily. These platforms allow users to have freedom of expression. However, some users exploit this facility by disseminating hate speeches. Manual detection and censorship of such hate speeches are impractical; thus, an automatic detection mechanism is required to detect and counter hate speeches in a real-time environment. Most research in hate speech detection has been carried out in the English language. Still, minimal work has been explored in other languages, mainly Urdu written in Roman Urdu script. A few research have attempted machine learning, and deep learning models for Roman Urdu hate speech detection; however, due to a scarcity of Roman Urdu resources, and a large corpus with defined annotation rules, a robust hate speech detection model is still required. With this motivation, this study contributes in the following manner: we developed annotation guidelines for Roman Urdu Hate Speech. Second, we constructed a new Roman Urdu Hate Speech Dataset (RU-HSD-30K) that was annotated by a team of experts using the annotation rules. To the best of our knowledge, the Bi-LSTM model with an attention layer for Roman-Urdu Hate Speech Detection has not been explored. Therefore, we developed a context-aware Roman Urdu Hate Speech detection model based on Bi-LSTM with an attention layer and used custom word2vec for word embeddings. Finally, we examined the effect of lexical normalization of Roman Urdu words on the performance of the proposed model. Different traditional as well as deep learning models, including LSTM and CNN models, were used as baseline models. The performance of the models was assessed in terms of evaluation matrices like accuracy, precision, recall, and F1-score. The generalization of each model is also evaluated on a cross-domain dataset. Experimental results revealed that Bi-LSTM with attention outperformed the traditional machine learning models and other deep learning models with an accuracy score of 0.875 and an F-Score of 0.885. In addition, the results demonstrated that our suggested model (Bi-LSTM with Attention Layer) is more general than previous models when applied to unseen data. The results confirmed that lexical normalization of Roman Urdu words enhanced the performance of the suggested model.

Monitoring insect activity at hive entrances is essential for advancing precision beekeeping practices by enabling non-invasive, real-time assessment of the colony’s health and early detection of potential threats. This study evaluates deep learning models for detecting worker bees, pollen-bearing bees, drones, and wasps, comparing different YOLO-based architectures optimized for real-time inference on an RTX 4080 Super and Jetson AGX Orin. A new publicly available dataset with diverse environmental conditions was used for training and validation. Performance comparisons showed that modified YOLOv8 models achieved a better precision–speed trade-off relative to other YOLO-based architectures, enabling efficient deployment on embedded platforms. Results indicate that model modifications enhance detection accuracy while reducing inference time, particularly for small object classes such as pollen. The study explores the impact of different annotation strategies on classification performance and tracking consistency. The findings demonstrate the feasibility of deploying AI-powered hive monitoring systems on embedded platforms, with potential applications in precision beekeeping and pollination surveillance.

On-road car accidents are immensely unfortunate but quite common occurrences worldwide. Instant data-centric and informed decisions of crisis management are rarely experienced due to the absence of real-time car accident detection and severity analysis mechanisms. On this background, the current paper presents a deep learning model for car accident detection and analysis of its severity so that the crisis management activities might follow without any delay saving invaluable human lives. The existing works lack in using time-series data, the proper learning model for accurate prediction, and minimizing the time taken in post-accident scenarios for the victims to receive immediate medical help. This paper introduces the Long Short Term Memory (LSTM) model in conjunction with the Gradient Boosted Regression Trees (GBRT) technique for the determination of car accidents with different levels of severity. The proposed model works with the accelerometer and gyroscopic data collected through an application installed in the smartphones of the users inside the car. The LSTM-GBRT hybrid model is proposed to achieve higher accuracy than LSTM which deals with time-variant data. The satisfactory performance of the proposed technique has been reported and the results are extensively investigated in comparison with another hybrid technique such as LSTM with Random Forest (RF) as well. The statistics confirm the superiority of the proposed model over other parallel models in terms of several performance metrics, like Accuracy, Precision, etc.

Respiratory-related leg movements (RRLMs) may contribute to the cardiovascular risk associated with obstructive sleep apnea (OSA). Selective serotonin reuptake inhibitors (SSRIs), but not bupropion, increase periodic leg movements in sleep. This study examines whether patients with OSA using SSRIs have more RRLMs than those taking bupropion or no antidepressant. Patients with an apnea-hypopnea index (AHI) of at least 10 events/h during a full-night diagnostic study or split-night study, who were taking bupropion (n = 32), an SSRI (n = 31), or no antidepressant (n = 31), were selected from a database of prestudy questionnaires. RRLMs were scored according to World Association of Sleep Medicine 2016 standards. Patients using SSRIs had significantly greater overall RRLM% (defined as the percentage of respiratory events associated with a leg movement, including apneas, hypopneas, and respiratory effort-related arousals), RRLM index, and periodic limb movement index relative to patients using bupropion and control patients. The difference between the RRLM% in the SSRI and bupropion groups was limited to patients undergoing split-night studies, and that of the SSRI and control groups was limited to patients undergoing full-night diagnostic studies. The greater number of RRLMs and PLMs in the SSRI group may contribute to treatment-emergent insomnia often seen with SSRI use. Fragmented sleep and elevated autonomic nervous system activation associated with increased RRLMs in patients with OSA taking SSRIs might also limit the tolerability of antidepressant treatment, as well as increase the risk for cardiovascular disease.

The aim of the study was to inspect the acoustic properties and sleep characteristics of a preapneic snoring sound. The feasibility of forecasting upcoming respiratory events by snoring sound was also investigated. Participants with habitual snoring or a heavy breathing sound during sleep were recruited consecutively. Polysomnography was conducted, and snoring-related breathing sound was recorded simultaneously. Acoustic features and sleep features were extracted from 30-second samples, and a machine learning algorithm was used to establish 2 prediction models. A total of 74 eligible participants were included. Model 1, tested by 5-fold cross-validation, achieved an accuracy of 0.92 and an area under the curve of 0.94 for respiratory event prediction. Model 2, with acoustic features and sleep information tested by Leave-One-Out cross-validation, had an accuracy of 0.78 and an area under the curve of 0.80. Sleep position was found to be the most important among all sleep features contributing to the performance of the 2 models. Preapneic sound presented unique acoustic characteristics, and snoring-related breathing sound could be deployed as a real-time apneic event predictor. The models, combined with sleep information, serve as a promising tool for an early warning system to forecast apneic events. Wang B, Yi X, Gao J, etal. Real-time prediction of upcoming respiratory events via machine learning using snoring sound signal. J Clin Sleep Med. 2021;17(9):1777-1784.

The purpose of this study is to conduct a systematic review and meta-analysis evaluating the effects of respiratory muscle therapy (ie, oropharyngeal exercises, speech therapy, breathing exercises, wind musical instruments) compared with control therapy or no treatment in improving apnea-hypopnea index ([AHI] primary outcome), sleepiness, and other polysomnographic outcomes for patients diagnosed with obstructive sleep apnea (OSA). Only randomized controlled trials with a placebo therapy or no treatment searched using PubMed, EMBASE, Cochrane, and Web of Science up to November 2018 were included, and assessment of risk of bias was completed using the Cochrane Handbook. Nine studies with 394 adults and children diagnosed with mild to severe OSA were included, all assessed at high risk of bias. Eight of the 9 studies measured AHI and showed a weighted average overall AHI improvement of 39.5% versus baselines after respiratory muscle therapy. Based on our meta-analyses in adult studies, respiratory muscle therapy yielded an improvement in AHI of -7.6 events/h (95% confidence interval [CI] = -11.7 to -3.5; P ≤ .001), apnea index of -4.2 events/h (95% CI = -7.7 to -0.8; P ≤ .016), Epworth Sleepiness Scale of -2.5 of 24 (95% CI= -5.1 to -0.1; P ≤ .066), Pittsburgh Sleep Quality Index of -1.3 of 21 (95% CI= -2.4 to -0.2; P ≤ .026), snoring frequency (P = .044) in intervention groups compared with controls. This systematic review highlights respiratory muscle therapy as an adjunct management for OSA but further studies are needed due to limitations including the nature and small number of studies, heterogeneity of the interventions, and high risk of bias with low quality of evidence.

Many patients with severe obstructive sleep apnea (OSA) do not complain of excessive daytime sleepiness (EDS), possibly due to increased sympathetic nervous activity (SNA) and accompanying heightened alertness. We hypothesized that in patients with OSA, those without subjective EDS (Epworth Sleepiness Scale, ESS score < 11) would have higher very low frequency (VLF) heart rate variability (HRV) during sleep, reflecting greater sympathetic heart rate modulation than patients with an ESS score ≥ 11. Patients with severe OSA (AHI ≥ 30: 26 with and 65 without heart failure) were divided into those with and without EDS. Heart rate (HR) signals were acquired in stage 2 sleep during periods of recurrent apneas and hypopneas and submitted to coarse graining spectral analysis, which extracts harmonic, neurally mediated contributions to HRV from total spectral power. Because the apnea-hyperpnea cycle entrains muscle SNA at VLF (0 to 0.04 Hz), VLF power was our principal between-group comparison. Subjects without EDS had higher harmonic VLF power (944 ± 839 vs 447 ± 461 msec(2), p = 0.003) than those with EDS, irrespective of the presence or absence of heart failure (1218 ± 944 vs 426 ± 299 msec(2), p = 0.043, and 1029 ± 873 vs 503 ± 533 msec(2), p = 0.003, respectively). ESS scores correlated inversely with VLF power in all (r = -0.294, p = 0.005) and in heart failure subjects (r = -0.468, p = 0.016). Patients with severe OSA but without EDS have higher VLF-HRV than those with EDS. This finding suggests that patients with severe OSA but without EDS have greater sympathetic modulation of HRV than those with EDS that may reflect elevated adrenergically mediated alertness. Taranto Montemurro L; Floras JS; Picton P; Kasai T; Alshaer H; Gabriel JM; Bradley TD. Relationship of heart rate variability to sleepiness in patients with obstructive sleep apnea with and without heart failure.

Cardiovascular Disease and Sleep Disordered Breathing: Are Children Vulnerable?

Obstructive Sleep Apnea (OSA) is a sleep disorder with a high prevalence in the general population. It is a risk factor for many cardiovascular diseases, and an independent risk factor for cerebrovascular diseases such as stroke. After an apnea episode, both arterial blood pressure and cerebral blood flow velocity change in function of the apnea duration (AD). We hypothesized that the relative excursion in heart rate (AHR), defined as the percentage difference between the maximum and the minimum heart rate values associated to an obstructive apnea event, is also related to AD. In this work we studied the relationship between apnea-related AHR and AD in a population of eight patients with severe OSA. AHR and AD showed a moderate but statistically significant correlation (p <; 0.0001) in a total of 1454 obstructive apneas analyzed. The average heart rate excursion for apneas with AD ≥ 30s (ΔHR = 31.29 ± 6.64%) was significantly greater (p = 0.0002) than for apneas with AD ∈ [10,20)s (ΔHR = 18.14±3.08%). We also observed that patients with similar Apnea-Hypopnea Index (AHI) may exhibit remarkably different distributions of AHR and AD, and that patients with a high AHI need not have a higher average AHR than others with a lower severity index. We conclude that the overall apnea-induced heart rate excursion is partially explained by the duration of apnoeic episodes, and it may be a simple measure of the cardiovascular stress associated with OSA that is not directly reflected in the AHI.

Both obstructive sleep apnea syndrome (OSAS) and sleep bruxism (SB) are commonly related to arousal events. In this study, we examined the effect of SB on the sleep architecture and investigated the relationship between SB and sleep respiratory events in patients with OSAS. Patients with OSAS (n=67) in whom apnea/hypopnea occurred five or more times per hour were recruited to this study. Healthy volunteers (n=16) were recruited as controls. None of the healthy volunteers had any sleep disorders or medical disorders, nor had they taken any medication or alcohol. Data were collected by standard polysomnography during overnight sleep tests in a dark, quiet room. The frequency of SB was higher in the OSAS than in the control group. The risk of SB was significantly higher in the OSAS than in the control group (odds ratio, 3.96; 95% confidence interval, 1.03-15.20; P<0.05). Apnea/hypopnea and desaturation events occurred significantly more frequently in patients with than without SB. The frequency of the phasic type of SB correlated positively with that of obstructive apnea, micro-arousal, and oxygen desaturation. The frequency of SB events during micro-arousal events consequent on apnea/hypopnea events was significantly higher in the OSAS than in the control group. We found that patients with OSAS have a high risk of SB. In particular, this is the first report relating phasic-type SB to obstructive apnea events. This relationship suggests that improvement in OSAS might prevent exacerbations of SB.

Primary Snoring in Children—No Longer Benign

<h3>Background</h3> Hypertrophic cardiomyopathy (HCM) represents the most common inherited cardiomyopathy characterized by otherwise unexplained left ventricular hypertrophy. Sleep disordered breathing (SDB) (including both central and obstructive sleep apnea) is considered to be an important and potentially reversible cause of cardiovascular disease progression. This prospective trial aims to define the prevalence of SDB within patients diagnosed with HCM using gold standard polysomnography (PSG). Previous studies have suggested a prevalence of SDB in the general population of 25.5%. <h3>Methods</h3> Previous trials have been retrospective studies and have used overnight oximetry analysis to find the prevalence of SDB in HCM using portable, at-home monitors, however overnight oximetry is unable to accurately delineate between central and obstructive sleep apnea – both of which vary significantly in terms of prognosis and management - and is not as sensitive nor specific in diagnosing or classifying SDB as polysomnography assessment. An ongoing prospective analysis of 85 patients diagnosed with HCM was performed at Mayo Clinic using PSG. Apnea is defined as the absence of inspiratory airflow for at least 10 seconds. Hypopnea is defined as a decrease in airflow lasting 10 seconds or longer associated with a desaturation &gt;4%. Apnea Hypopnea Index (AHI) – the number of events per hour was assessed using PSG. SDB is defined as an AHI&gt;5/hour of sleep. <h3>Results</h3> Of 85 HCM patients examined using PSG, 49 were found to have an AHI&gt;5/hour. Average AHI was 20.0, Standard Deviation 23.3, Interquartile range 3.3-28.2. 15 patients had central sleep apnea defined as the absence of inspiratory effort for at least 10 seconds. 17 patients had obstructive apneas. 18 patients had severe SDB (AHI&gt;30/hour). <h3>Conclusion</h3> Given the paucity of available treatments for HCM and noting that treatment of SDB in heart failure with CPAP has been shown to attenuate diastolic dysfunction, this represents a potentially novel avenue of treatment for patients diagnosed with HCM. OSA is known to increase risk factors that contribute to morbidity and mortality in HCM including arrhythmias, myocardial infarction and sudden cardiac death. Previous studies have suggested a higher prevalence of non-sustained ventricular tachycardia in HCM patients diagnosed with OSA likely explained by excessive sympathetic activation. Systemic blood pressure and sympathetic activity have also previously been seen to improve with CPAP therapy and there is evidence to suggest CPAP therapy may reverse or slow progression of left ventricular hypertrophy in HCM patients with SDB. Our study suggests a significantly higher prevalence of SDB in HCM patients when compared to the general population and whether treatment of SDB in HCM improves outcomes requires further investigation which our prospective study aims to answer. <h3>Conflict of Interest</h3> None

Objective: To investigate the characteristics of sleep-related respiratory events in normal children and to provide normal polysomnographic parameters for diagnosing sleep-disordered breathing in children. Methods: Normal subjects between 3 and 14 years old were enrolled from 1 July 2014 to 31 December 2015 and the subjects received overnight polysomnography at the sleep center of our hospital. They were children of our hospital employees or were recruited from the communities who did not have sleep and respiratory disorders. The children were divided into preschool group (3-5 years) and school-age group (6-14 years). Apnea index (AI), obstructive apnea index (OAI), central apnea index (CAI), and mixed apnea index (MAI) were compared between the two groups. Data for continuous variables that showed normal distribution were expressed as x ±s. M(P25, P75) were used when data were not normally distributed. Continuous variables that showed normal distribution were compared by using an independent-sample t-test. Wilcoxon-test was performed when data exhibited non-normal distribution. Differences in categorical data were tested with Chi-square test. Pearson correlation test was applied for the correlation analysis. P<0.05 was considered statistically significant. Results: A total of 115 normal children took part in the study including 40 in preschool group and 75 in school-age group. Children in both groups had a few sleep apnea events, most of which were central apneas, accounting for 80% and 70% of the total respiratory events respectively. Central apnea index in preschool children were significantly higher than that of school-age children (P<0.001), with median of 0.6 times/h and 0.1 times/h, respectively. Median OAI of both groups were 0.0 times/h without significant difference (P=0.748). Obstructive apnea events occurred mainly in the supine position in both groups. Conclusions: Normal children may have a few apnea events in sleep that were predominantly central apnea. CAI of preschool children is significantly higher than that of school-age children. Obstructive sleep apnea is rare in normal children, and sleep apnea occurs mainly in the supine position.

Computed tomography has been used to study the pharyngeal airway during tidal breathing in wakefulness and during obstructive apnoeas in Non-REM sleep in patients with obstructive sleep apnoea. In supine subjects, contiguous transverse 10 mm sections were taken perpendicular to the posterior pharyngeal wall with a 2.1 s scan time. Studies during wakefulness showed that the narrowest section of the pharyngeal airspace was in the region posterior to the soft palate and that the minimal airway cross-sectional areas were significantly reduced in the group of patients with obstructive sleep apnoea compared to the group of control subjects without obstructive sleep apnoea. The studies during sleep showed that in all patients, the airspace posterior to the soft palate was a site of obstructive apnoeas. The length of the obstructed segment varied between patients, extending below the level of the soft palate in half the patient group. Airway narrowing and obstruction was due to posterior displacement of the soft palate and the tongue in the majority of patients, although lateral displacement of the pharyngeal walls was also observed. No occlusion was observed in the laryngopharynx although there was narrowing of oro- and laryngopharyngeal apertures below the site of obstruction during obstructive apnoeas. The size of the oropharyngeal airspace during wakefulness did not predict the presence of airway occlusion below the level of the soft palate when asleep. The variability between patients in the site(s) of upper airway obstruction during obstructive apnoeas have important implications for the choice of appropriate treatment in patients with obstructive sleep apnoea.

Obstructive sleep apnea causes intermittent elevation of systemic blood pressure (BP) during sleep. To determine whether obstructive apnea in children has a tonic effect on diurnal BP, 24-hour ambulatory blood pressure was obtained from 60 children with mean age of 10.8 +/- 3.5 years. Thirty-nine children had obstructive apnea and 21 had primary snoring. Children with obstructive apnea had significantly greater mean BP variability during wakefulness and sleep, a higher night-to-day systolic BP, and a smaller nocturnal dipping of mean BP. Variability of mean arterial pressure during wakefulness was predicted by the desaturation, body mass, and arousal indices, whereas variability during sleep was predicted by apnea-hypopnea and body mass indices. Nocturnal BP dipping was predicted by the desaturation index. There were no significant differences in systolic, diastolic, or mean arterial BP during sleep between the groups. Diastolic BP during wakefulness was significantly different between the groups and correlated negatively with apnea-hypopnea index. We conclude that obstructive apnea in children is associated with 24-hour BP dysregulation and that, independent of obesity, the frequency of obstructive apnea, oxygen desaturation, and arousal contributes to abnormal BP control.