Automating sleep staging is vital to scale up sleep assessment and diagnosis to serve millions experiencing sleep deprivation and disorders and enable longitudinal sleep monitoring in home environments. Learning from raw polysomnography signals and their derived time-frequency image representations has been prevalent. However, learning from multi-view inputs (e.g., both the raw signals and the time-frequency images) for sleep staging is difficult and not well understood. This work proposes a sequence-to-sequence sleep staging model, XSleepNet, ¹ that is capable of learning a joint representation from both raw signals and time-frequency images. Since different views may generalize or overfit at different rates, the proposed network is trained such that the learning pace on each view is adapted based on their generalization/overfitting behavior. In simple terms, the learning on a particular view is speeded up when it is generalizing well and slowed down when it is overfitting. View-specific generalization/overfitting measures are computed on-the-fly during the training course and used to derive weights to blend the gradients from different views. As a result, the network is able to retain the representation power of different views in the joint features which represent the underlying distribution better than those learned by each individual view alone. Furthermore, the XSleepNet architecture is principally designed to gain robustness to the amount of training data and to increase the complementarity between the input views. Experimental results on five databases of different sizes show that XSleepNet consistently outperforms the single-view baselines and the multi-view baseline with a simple fusion strategy. Finally, XSleepNet also outperforms prior sleep staging methods and improves previous state-of-the-art results on the experimental databases.

Human sleep is cyclical with a period of approximately 90 minutes, implying long temporal dependency in the sleep data. Yet, exploring this long-term dependency when developing sleep staging models has remained untouched. In this work, we show that while encoding the logic of a whole sleep cycle is crucial to improve sleep staging performance, the sequential modelling approach in existing state-of-the-art deep learning models are inefficient for that purpose. We thus introduce a method for efficient long sequence modelling and propose a new deep learning model, L-SeqSleepNet, which takes into account whole-cycle sleep information for sleep staging. Evaluating L-SeqSleepNet on four distinct databases of various sizes, we demonstrate state-of-the-art performance obtained by the model over three different EEG setups, including scalp EEG in conventional Polysomnography (PSG), in-ear EEG, and around-the-ear EEG (cEEGrid), even with a single EEG channel input. Our analyses also show that L-SeqSleepNet is able to alleviate the predominance of N2 sleep (the major class in terms of classification) to bring down errors in other sleep stages. Moreover the network becomes much more robust, meaning that for all subjects where the baseline method had exceptionally poor performance, their performance are improved significantly. Finally, the computation time only grows at a sub-linear rate when the sequence length increases.

Background: Cardiac output (Q)over dot monitoring can support the management of high-risk surgical patients, but the pulmonary artery catheterisation required by the current 'gold standard'-bolus thermodilution ((Q)over dot(T))-has the potential to cause life-threatening complications. We present a novel noninvasive and fully automated method that uses the inspired sinewave technique to continuously monitor cardiac output ((Q)over dot(IsT)). Methods: Over successive breaths the inspired nitrous oxide (N2O) concentration was forced to oscillate sinusoidally with a fixed mean (4%), amplitude (3%), and period (60 s). (Q)over dot(IST) was determined in a single-compartment tidal ventilation lung model that used the resulting amplitude/phase of the expired N2O sinewave. The agreement and trending ability of (Q)over dot(IST) were compared with (Q)over dot(T) during pharmacologically induced haemodynamic changes, before and after repeated lung lavages, in eight anaesthetised pigs. Results: Before lung lavage, changes in (Q)over dot(IST) and (Q)over dot(T) from baseline had a mean bias of -0.52 L min(-1) (95% confidence interval [CI], -0.41 to -0.63). The concordance between (Q)over dot(IST) and (Q)over dot(T) was 92.5% as assessed by four-quadrant analysis, and polar plot analysis revealed a mean angular bias of 5.98 degrees (95% CI, -24.4 degrees-36.3 degrees). After lung lavage, concordance was slightly reduced (89.4%), and the mean angular bias widened to 21.8 degrees (-4.2 degrees, 47.6 degrees). Impaired trending ability correlated with shunt fraction (r=0.79, P<0.05). Conclusions: The inspired sinewave technique provides continuous and noninvasive monitoring of cardiac output, with a 'marginal-good' trending ability compared with cardiac output based on thermodilution. However, the trending ability can be reduced with increasing shunt fraction, such as in acute lung injury.

Bedside lung volume measurement could personalise ventilation and reduce driving pressure in patients with acute respiratory distress syndrome (ARDS). We investigated a modified gas-dilution method, the inspired sinewave technique (IST), to measure the effective lung volume (ELV) in pigs with uninjured lungs and in an ARDS model. Anaesthetised mechanically ventilated pigs were studied before and after surfactant depletion by saline lavage. Changes in PEEP were used to change ELV. Paired measurements of absolute ELV were taken with IST (ELV_IST) and compared with gold-standard measures (sulphur hexafluoride wash in/washout [ELV_SF6] and computed tomography (CT) [ELV_CT]). Measured volumes were used to calculate changes in ELV (ΔELV) between PEEP levels for each method (ΔELV_IST, ΔELV_SF6, and ΔELV_CT). The coefficient of variation was <5% for repeated ELV_IST measurements (n=13 pigs). There was a strong linear relationship between ELV_IST and ELV_SF6 in uninjured lungs (r²=0.97), and with both ELV_SF6 and ELV_CT in the ARDS model (r²=0.87 and 0.92, respectively). ELV_IST had a mean bias of -12 to 13% (95% limits=±17 - 25%) compared with ELV_SF6 and ELV_CT. ΔELV_IST was concordant with ΔELV_SF6 and ΔELV_CT in 98-100% of measurements, and had a mean bias of -73 to -77 ml (95% limits=±128 - 186 ml) compared with ΔELV_SF6 and -1 ml (95% limits ±333 ml) compared with ΔELV_CT. IST provides a repeatable measure of absolute ELV and shows minimal bias when tracking PEEP-induced changes in lung volume compared with CT in a saline-lavage model of ARDS.

Background: Direct oral anticoagulants (DOACs) are the standard treatment for stroke prevention in AF. However, high-quality head-to-head comparisons of DOACs are lacking. This study compared oral anticoagulants in patients with AF. Methods: Data were retrieved from eligible randomised controlled trials (RCTs). Interventions were ranked using the surface under the cumulative ranking curve (SUCRA) and the frequentist random effects model was applied. Efficacy outcomes included stroke, systemic embolism, MI, and all-cause mortality; the safety outcome was major bleeding. A composite outcome of efficacy and net clinical benefit was also evaluated. Results: From 23,152 records, 11 eligible RCTs were identified and included in the study. Rivaroxaban was superior to vitamin K antagonists (VKA) in net clinical benefit (RR 0.75; 95% CI [0.59–0.94]; p=0.0133), but there were no significant differences between other DOACs and VKA or among the DOACs themselves. Rivaroxaban reduced the risk of the composite outcome of efficacy compared with dabigatran (RR 0.85; 95% CI [0.75–0.98]; p=0.02) and edoxaban (RR 0.84; 95% CI [0.75–0.95]; p=0.0051), but not apixaban (RR 0.89; 95% CI [0.89–1.02]; p=0.087). All DOACs showed superiority over VKA in efficacy, without an increased risk of major bleeding. Based on the SUCRA, rivaroxaban showed a favourable risk–benefit profile compared with the other anticoagulants. Conclusion: This study showed that DOACs are superior to VKA in efficacy without increasing major bleeding risk, with rivaroxaban demonstrating the most balanced risk–benefit profile. Well-designed RCTs are needed to validate these findings.

The prevalence and predictors of mortality following an ischemic stroke or intracerebral hemorrhage have not been well established among patients in Vietnam. 2885 consecutive diagnosed patients with ischemic stroke and intracerebral hemorrhage at ten stroke centres across Vietnam were involved in this prospective study. Posthoc analyses were performed in 2209 subjects (age was 65.4 ± 13.7 years, with 61.4% being male) to explore the clinical characteristics and prognostic factors associated with 90-day mortality following treatment. An explainable machine learning model using extreme gradient boosting and SHapley Additive exPlanations revealed the correlation between original clinical research and advanced machine learning methods in stroke care. In the 90 days following treatment, the mortality rate for ischemic stroke was 8.2%, while for intracerebral hemorrhage, it was higher at 20.5%. Atrial fibrillation was an elevated risk of 90-day mortality in the ischemic stroke patient (OR 3.09; 95% CI 1.90–5.02, p 0.05). The baseline NIHSS score was a significant predictor of 90-day mortality in both patient groups. The machine learning model can predict a 0.91 accuracy prediction of death rate after 90 days. Age and NIHSS score were in the top high risks with other features, such as consciousness, heart rate, and white blood cells. Stroke severity, as measured by the NIHSS, was identified as a predictor of mortality at discharge and the 90-day mark in both patient groups.