Deep transfer learning for automated single-lead EEG sleep staging with channel and population mismatches

Frontiers in Physiology, Год журнала: 2024, Номер 14

Опубликована: Янв. 5, 2024

Introduction: Automated sleep staging using deep learning models typically requires training on hundreds of recordings, and pre-training public databases is therefore common practice. However, suboptimal stage performance may occur from mismatches between source target datasets, such as differences in population characteristics (e.g., an unrepresented disorder) or sensors alternative channel locations for wearable EEG). Methods: We investigated three strategies automated single-channel EEG stager: (i.e., the original dataset), training-from-scratch new fine-tuning dataset, dataset). As we used F3-M2 healthy subjects (N = 94). Performance different was evaluated Cohen’s Kappa ( κ ) eight smaller datasets consisting 60), patients with obstructive apnea (OSA, N insomnia REM behavioral disorder (RBD, 22), combined two channels, F3-F4. Results: No observed age-matched average across .83 healthy, .77 insomnia, .74 OSA subjects. RBD set, where data availability limited, preferred method .67), increase .15 to training-from-scratch. In presence mismatches, targeted required, either through fine-tuning, increasing .17 average. Discussion: found that, when and/or cause performance, a approach can yield similar superior compared building model scratch, while requiring sample size. contrast OSA, contains characteristics, inherent pathology age-related, which apparently demand training.

Язык: Английский

---

Hypnogram and Hypnodensity Analysis of REM Sleep Behaviour Disorder Using Both EEG and HRV‐Based Sleep Staging Models

Journal of Sleep Research, Год журнала: 2025, Номер unknown

Опубликована: Март 12, 2025

ABSTRACT Rapid‐eye‐movement (REM) sleep behaviour disorder (RBD) is a primary strongly associated with Parkinson's disease. Assessing structure in RBD important for understanding the underlying pathophysiology and developing diagnostic methods. However, performance of automated stage classification (ASSC) models considered suboptimal RBD, both utilising neurological signals (“ExG”: EEG, EOG, chin EMG) heart rate variability combined body movements (HRVm). Here, we explore this underperformance through categorical representation macrostructure (i.e., hypnogram) that leverages probability distribution ASSCs hypnodensity). By comparing population ( n = 36) to sex‐ age‐matched group OSA patients chosen their anticipated similarly decreased stability, confirm lower 4‐stage ExG‐based ASSC (RBD: κ 0.74, OSA: 0.80) HRVm‐based 0.50, 0.63). Stages showing agreement namely, N1 + N2 REM sleep, exhibited elevated ambiguity hypnodensity, indicating more ambiguous distributions. Limited differences bout durations between suggested instability not necessarily driving RBD. transitions showed abrupt changes distribution, while had continuous profile, possibly complicating classification. Although staging remain challenging, hypnodensity analysis informative characterisation can capture potential drivers disagreement.

Язык: Английский

---

Camera-Based Continuous Heart and Respiration Rate Monitoring in the ICU

Applied Sciences, Год журнала: 2025, Номер 15(7), С. 3422 - 3422

Опубликована: Март 21, 2025

We provide new insights into the performance of camera-based heart and respiration rate extraction evaluate its usability for replacing spot checks conducted in general ward. A study was performed comprising 36 ICU patients recorded a total time 699 h. The 5 beats/minute agreement between camera ECG-based measurements 81.5%, with coverage 81.9%, where largest gap 239 min. challenges encountered monitoring were limited visibility patient’s face irregular rates, which led to poor camera- measurements. To prevent non-breathing motion from causing error extraction, we developed metric used detect motion. 3 breaths/minute contact-based 91.1%, 59.1%, 114 Encountered morphology signal breathing. While few need be overcome, results show promise as replacement these vital parameters

Язык: Английский

---

Feasibility of Radar Vital Sign Monitoring Using Multiple Range Bin Selection

Sensors, Год журнала: 2025, Номер 25(8), С. 2596 - 2596

Опубликована: Апрель 20, 2025

Radars are promising tools for contactless vital sign monitoring. As a screening device, radars could supplement polysomnography, the gold standard in sleep medicine. When radar is placed lateral to person, signs can be extracted simultaneously from multiple body parts. Here, we present method select every available breathing and heartbeat signal, instead of selecting only one optimal signal. Using concurrent signals enhance rate robustness accuracy. We built an algorithm based on persistence diagrams, modern tool time series analysis field topological data analysis. Multiple criteria were evaluated diagrams detect signals. tested feasibility simultaneous overnight polysomnography recordings six healthy participants. Compared against single bin selection, selection lead improved accuracy both (mean absolute error: 0.29 vs. 0.20 breaths per minute) heart 1.97 0.66 beats minute). Additionally, fewer artifactual segments selected. Furthermore, distribution chosen along aligned with basic physiological assumptions. In conclusion, monitoring benefit achieved by selection. The provide additional information

Язык: Английский

---

A computationally efficient algorithm for wearable sleep staging in clinical populations

Scientific Reports, Год журнала: 2023, Номер 13(1)

Опубликована: Июнь 6, 2023

This study describes a computationally efficient algorithm for 4-class sleep staging based on cardiac activity and body movements. Using an accelerometer to calculate gross movements reflective photoplethysmographic (PPG) sensor determine interbeat intervals corresponding instantaneous heart rate signal, neural network was trained classify between wake, combined N1 N2, N3 REM in epochs of 30 s. The classifier validated hold-out set by comparing the output against manually scored stages polysomnography (PSG). In addition, execution time compared with that previously developed variability (HRV) feature-based algorithm. With median epoch-per-epoch κ 0.638 accuracy 77.8% achieved equivalent performance when HRV-based approach, but 50-times faster time. shows how network, without leveraging any priori knowledge domain, can automatically "discover" suitable mapping movements, stages, even patients different pathologies. addition high performance, reduced complexity makes practical implementation feasible, opening up new avenues diagnostics.

Язык: Английский

---

Feasibility of Exploiting Physiological and Motion Features for Camera-based Sleep Staging: A Clinical Study

2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), Год журнала: 2023, Номер unknown, С. 1 - 5

Опубликована: Июль 24, 2023

Camera-based sleep monitoring is an emergent research topic in medicine. The feasibility of using both the physiological features and motion measured by a video camera for staging was not thoroughly investigated. In this paper, we built camera-based non-contact setup Institute Respiratory Diseases, Shenzhen People's Hospital, created clinical dataset (nocturnal data 11 adults) including expert-corrected PSG references synchronized with video. measurements have shown high correlations PSG. It obtains overall Mean Absolute Error (MAE) 1.5 bpm heart-rate (HR), 0.7 breathing-rate (BR), 13.9 ms variability (HRV), accuracy 93.5% leg detection. statistical analysis indicates that averaged HR variations BR are distinct annotating four stages (awake, REM, light sleep, deep sleep). HRV parameter (SDNN) can clearly differentiate rapid-eye-movement (REM) non-REM, while movement distinctive feature separating awake sleep. trial demonstrated joint staging, provides insights sleep-related selection.

Язык: Английский

---

Deep Learning-Enabled Sleep Staging From Vital Signs and Activity Measured Using a Near-Infrared Video Camera

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Год журнала: 2023, Номер unknown

Опубликована: Июнь 1, 2023

Conventional sleep monitoring is time-consuming, expensive and uncomfortable, requiring a large number of contact sensors to be attached the patient. Video data commonly recorded as part laboratory assessment. If accurate staging could achieved solely from video, this would overcome many problems traditional methods. In work we use heart rate, breathing rate activity measures, all derived near-infrared video camera, perform stage classification. We deep transfer learning approach scarcity, by using an existing contact-sensor dataset learn effective representations time series. Using 50 healthy volunteers, achieve accuracy 73.4% Cohen's kappa 0.61 in four-class classification, establishing new state-of-the-art for video-based staging.

Язык: Английский

---

The impact of healthy pregnancy on features of heart rate variability and pulse wave morphology derived from wrist-worn photoplethysmography

Scientific Reports, Год журнала: 2023, Номер 13(1)

Опубликована: Ноя. 30, 2023

Due to the association between dysfunctional maternal autonomic regulation and pregnancy complications, tracking non-invasive features of derived from wrist-worn photoplethysmography (PPG) measurements may allow for early detection deteriorations in health. However, even though a plethora these features-specifically, describing heart rate variability (HRV) morphology PPG waveform (morphological features)-exist literature, it is unclear which be valuable As an initial step towards clarity, we compute comprehensive sets HRV morphological nighttime measurements. From these, using logistic regression stepwise forward feature elimination, identify that best differentiate healthy pregnant women non-pregnant women, since likely capture physiological adaptations necessary sustaining pregnancy. Overall, were more discriminating than (area under receiver operating characteristics curve 0.825 0.74, respectively), with systolic pulse wave deterioration being most single feature, followed by mean (HR). Additionally, stratified analysis sleep stages found calculated only periods deep enhanced differences two groups. In conclusion, postulate addition features, also useful health suggest specific included future research concerning

Язык: Английский

---

Temporal dynamics of awakenings from slow‐wave sleep in non‐rapid eye movement parasomnia

Journal of Sleep Research, Год журнала: 2023, Номер 33(3)

Опубликована: Дек. 9, 2023

Summary Non‐rapid eye movement parasomnia disorders, also called disorders of arousal, are characterized by abnormal nocturnal behaviours, such as confusional arousals or sleep walking. Their pathophysiology is not yet fully understood, and objective diagnostic criteria lacking. It known, however, that behavioural episodes occur mostly in the beginning night, after an increase slow‐wave activity during sleep. A better understanding prospect may lead to new insights underlying mechanisms eventually facilitate diagnosis. We investigated temporal dynamics transitions from 52 patients 79 controls. Within patient group, non‐behavioural N3 awakenings were distinguished. Patients showed a higher probability wake up bout ended than controls, this increased with duration. Bouts longer 15 min resulted awakening 73% 34% time respectively. Behavioural reduced over cycles due reduction reducing ratio between awakenings. In first two cycles, bouts prior significantly shorter advancing patients, Our findings provide timing both N3, which result prediction potentially prevention episodes. This work, moreover, leads more complete characterization prototypical hypnogram parasomnias, could

Язык: Английский

---

Studying sleep: towards the identification of hypnogram features that drive expert interpretation

SLEEP, Год журнала: 2023, Номер 47(3)

Опубликована: Дек. 1, 2023

Hypnograms contain a wealth of information and play an important role in sleep medicine. However, interpretation the hypnogram is difficult task requires domain knowledge "clinical intuition." This study aimed to uncover which features drive by physicians. In other words, make explicit physicians implicitly look for hypnograms. Three experts evaluated up 612 hypnograms, indicating normal or abnormal structure suspicion disorders. ElasticNet convolutional neural network classification models were trained predict collected expert evaluations using stages as input. The several measures, including accuracy, Cohen's kappa, Matthew's correlation coefficient, confusion matrices. Finally, model coefficients visual analytics techniques used interpret associate with evaluation. Agreement between (Kappa 0.47 0.52) similar agreement 0.38 0.50). Sleep fragmentation, measured transitions per hour, stage distribution identified predictors interpretation. By comparing hypnograms not solely on epoch-by-epoch basis, but also these more specific that are relevant evaluation experts, performance assessment (automatic) sleep-staging surrogate trackers may be improved. particular, fragmentation feature deserves attention it often included PSG report, existing (wearable) have shown relatively poor this aspect.

Язык: Английский

---