OBJECTIVE:Rapid-Eye-Movement (REM) sleep behaviour disorder (RBD) is an early predictor of Parkinson's disease, dementia with Lewy bodies, and multiple system atrophy. This study investigated the use of a minimal set of sensors to achieve effective screening for RBD in the population, integrating automated sleep staging (three state) followed by RBD detection without the need for cumbersome electroencephalogram (EEG) sensors. METHODS:Polysomnography signals from 50 participants with RBD and 50 age-matched healthy controls were used to evaluate this study. Three stage sleep classification was achieved using a random forest classifier and features derived from a combination of cost-effective and easy to use sensors, namely electrocardiogram (ECG), electrooculogram (EOG), and electromyogram (EMG) channels. Subsequently, RBD detection was achieved using established and new metrics derived from ECG and EMG channels. RESULTS:The EOG and EMG combination provided the optimal minimalist fully-automated performance, achieving 0.57 ± 0.19 kappa (3 stage) for sleep staging and an RBD detection accuracy of 0.90 ± 0.11, (sensitivity and specificity of 0.88 ± 0.13 and 0.92 ± 0.098, respectively). A single ECG sensor achieved three state sleep staging with 0.28 ± 0.06 kappa and RBD detection accuracy of 0.62 ± 0.10. CONCLUSIONS:This study demonstrates the feasibility of using signals from a single EOG and EMG sensor to detect RBD using fully-automated techniques. SIGNIFICANCE:This study proposes a cost-effective, practical, and simple RBD identification support tool using only two sensors (EMG and EOG); ideal for screening purposes.

The severity of COVID-19 varies dramatically, ranging from asymptomatic infection to severe respiratory failure and death. Currently, few prognostic markers for disease outcomes exist, impairing patient triaging and treatment. Here, we train feed-forward neural networks on electronic health records of 819 confirmed SARS-CoV-2 positive patients admitted to a two-site NHS Trust hospital in London, England. To allow early risk assessment, the models ingest data collected in the emergency department (ED) to predict subsequent admission to intensive care, need for mechanical ventilation and in-hospital mortality. We apply univariate selection and recursive feature elimination to find the minimal subset of clinical variables needed for accurate prediction. Our models achieve AUC-ROC scores of 0.78 to 0.87, outperforming standard clinical risk scores. This accuracy is reached with as few as 13

Since its emergence in late 2019, the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has caused a pandemic with more than 55 million reported cases and 1.3 million estimated deaths worldwide. While epidemiological and clinical characteristics of COVID-19 have been reported, risk factors underlying the transition from mild to severe disease among patients remain poorly understood. In this retrospective study, we analysed data of 879 confirmed SARS-CoV-2 positive patients admitted to a two-site NHS Trust hospital in London, England, between January 1st and May 26th, 2020, with a majority of cases occurring in March and April. We extracted anonymised demographic data, physiological clinical variables and laboratory results from electronic healthcare records (EHR) and applied multivariate logistic regression, random forest and extreme gradient boosted trees. To evaluate the potential for early risk assessment, we used data available during patients’ initial presentation at the emergency department (ED) to predict deterioration to one of three clinical endpoints in the remainder of the hospital stay: admission to intensive care, need for invasive mechanical ventilation and in-hospital mortality. Based on the trained models, we extracted the most informative clinical features in determining these patient trajectories. Considering our inclusion criteria, we have identified 129 of 879 (15%) patients that required intensive care, 62 of 878 (7%) patients needing mechanical ventilation, and 193 of 619 (31%) cases of in-hospital mortality. Our models learned successfully from early clinical data and predicted clinical endpoints with high accuracy, the best model achieving area under the receiver operating characteristic (AUC-ROC) scores of 0.76 to 0.87 (F1 scores of 0.42–0.60). Younger patient age was associated with an increased risk of receiving intensive care and ventilation, but lower risk of mortality. Clinical indicators of a patient’s oxygen supply and selected laboratory results, such as blood lactate and creatinine levels, were most predictive of COVID-19 patient trajectories. Among COVID-19 patients machine learning can aid in the early identification of those with a poor prognosis, using EHR data collected during a patient’s first presentation at ED. Patient age and measures of oxygenation status during ED stay are primary indicators of poor patient outcomes.

Coronavirus disease (COVID-19), caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is rapidly spreading across the globe. The clinical spectrum of SARS-CoV-2 pneumonia requires early detection and monitoring, within a clinical environment for critical cases and remotely for mild cases, with a large spectrum of symptoms. The fear of contamination in clinical environments has led to a dramatic reduction in on-site referrals for routine care. There has also been a perceived need to continuously monitor non-severe COVID-19 patients, either from their quarantine site at home, or dedicated quarantine locations (e.g. hotels). In particular, facilitating contact tracing with proximity and location tracing apps was adopted in many countries very rapidly. Thus, the pandemic has driven incentives to innovate and enhance or create new routes for providing healthcare services at distance. In particular, this has created a dramatic impetus to find innovative ways to remotely and effectively monitor patient health status. In this paper, we present a review of remote health monitoring initiatives taken in 20 states during the time of the pandemic. We emphasize in the discussion particular aspects that are common ground for the reviewed states, in particular the future impact of the pandemic on remote health monitoring and consideration on data privacy.

Current management of bipolar disorder relies on self-reported questionnaires and interviews with clinicians. The development of objective measures of deteriorating mood may also allow for early interventions to take place to avoid transitions into depressive states. The objective of this study was to use acceleration data recorded from smartphones to predict levels of depression in a population of participants diagnosed with bipolar disorder. Data were collected from 52 participants, with a mean of 37 weeks of acceleration data with a corresponding depression score recorded per participant. Time varying hidden Markov models were used to extract weekly features of activity, sleep and circadian rhythms. Personalised regression achieved mean absolute errors of 1.00(0.57) from a possible scale of 0 to 27 and was able to classify depression with an accuracy of 0.84(0.16). The results demonstrate features derived from smartphone accelerometers are able to provide objective markers of depression. Low barriers for uptake exist due to the widespread use of smartphones, with personalised models able to account for differences in the behaviour of individuals and provide accurate predictions of depression.

In this work, we propose a multi-task recurrent neural network with attention mechanism for predicting cardiovascular events from electronic health records (EHRs) at different time horizons. The proposed approach is compared to a standard clinical risk predictor (QRISK) and machine learning alternatives using 5-year data from a NHS Foundation Trust. The proposed model outperforms standard clinical risk scores in predicting stroke (AUC=0.85) and myocardial infarction (AUC=0.89), considering the largest time horizon. Benefit of using an \gls{mt} setting becomes visible for very short time horizons, which results in an AUC increase between 2-6%. Further, we explored the importance of individual features and attention weights in predicting cardiovascular events. Our results indicate that the recurrent neural network approach benefits from the hospital longitudinal information and demonstrates how machine learning techniques can be applied to secondary care.

Determining phenotypes of diseases can have considerable benefits for in-hospital patient care and to drug development. The structure of high dimensional data sets such as electronic health records are often represented through an embedding of the data, with clustering methods used to group data of similar structure. If subgroups are known to exist within data, supervised methods may be used to influence the clusters discovered. We propose to extend deep embedded clustering to a semisupervised deep embedded clustering algorithm to stratify subgroups through known labels in the data. In this work we apply deep semi-supervised embedded clustering to determine data-driven patient subgroups of heart failure from the electronic health records of 4,487 heart failure and control patients. We find clinically relevant clusters from an embedded space derived from heterogeneous data. The proposed algorithm can potentially find new undiagnosed subgroups of patients that have different outcomes, and, therefore, lead to improved treatments.

Multi-source datasets often present the challenge of source-wise missing data which can render large portions of the dataset inaccessible. The applicability of traditional missing data techniques on multi-source datasets is poorly understood. We present the first quantitative evaluation of the state-of-the-art missing data techniques as applied to a freely available dataset of smart-phone recordings from Parkinsonian patients wherein source-wise missing data is simulated. The classification accuracy and imputation error of five missing data techniques, including a multi-modal autoencoder and multi-source ensemble learning, are compared at varying levels of missingness. These results demonstrate the relative applicability of each technique under different conditions and subsequently highlight the challenges of source-wise missing on remotely collected datasets. Specifically, multi-source ensemble learning proves to be a highly successful alternative to the traditional imputation techniques when a majority of observations possess missing data.

Reliable object detection is crucial for intelligent surveillance systems. Despite the abundance of available systems, these are often expensive or require cloud connectivity which restricts their usage. In this paper, we present a complete low-cost solution using embedded devices. Person detection is performed locally, using an efficient implementation of a state-of-the-art convolutional Single Shot Detector. The proposed model is fine-tuned to the surveillance task at hand and tested using a Raspberry Pi and the Up Squared Board devices. Additionally, the benefits of hardware-acceleration are evaluated using an Intel ® Movidius™ Neural Compute Stick (NCS). The final model performs person detection with an average precision of 0.48. The Raspberry Pi is capable of acquiring and processing images at a rate of 0.4 to 5.3 FPS without/with NCS, whereas the Up Board achieves 2.1 to 7.3 FPS. Further tests demonstrate how the system performs when multiple cameras and multiple NCS devices are deployed. In conclusion, the proposed architecture is suitable for real-time, affordable surveillance systems in home and commercial settings.

This study aims to develop automated diagnostic tools to aid in the identification of rapid-eye-movement (REM) sleep behaviour disorder (RBD). Those diagnosed with RBD enact their dreams and therefore present an abnormal characteristic of movement during REM sleep. Several methods have been proposed for RBD detection that use electromyogram (EMG) recordings and manually annotated sleep stages to objectively quantify abnormal REM movement. In this work we further develop these proven techniques with additional features that incorporate the relationship of muscle movement between sleep stages and general sleep architecture. Performance is evaluated using polysomnography (PSG) recordings from 43 aged-matched healthy controls and subjects diagnosed with RBD obtained from multiple institutions and publicly available resources. Using a random forest classifier with established and additional features, the performance of RBD detection was shown to improve upon established metrics (achieving 88% accuracy, 91% sensitivity, and 86% specificity).

Automatic sleep staging has been often treated as a simple classification problem that aims at determining the label of individual target polysomnography epochs one at a time. In this paper, we tackle the task as a sequence-to-sequence classification problem that receives a sequence of multiple epochs as input and classifies all of their labels at once. For this purpose, we propose a hierarchical recurrent neural network named SeqSleepNet (source code is available at http://github.com/pquochuy/SeqSleepNet). At the epoch processing level, the network consists of a filterbank layer tailored to learn frequency-domain filters for preprocessing and an attention-based recurrent layer designed for short-term sequential modeling. At the sequence processing level, a recurrent layer placed on top of the learned epoch-wise features for long-term modeling of sequential epochs. The classification is then carried out on the output vectors at every time step of the top recurrent layer to produce the sequence of output labels. Despite being hierarchical, we present a strategy to train the network in an end-to-end fashion. We show that the proposed network outperforms the state-of-the-art approaches, achieving an overall accuracy, macro F1-score, and Cohen’s kappa of 87.1%, 83.3%, and 0.815 on a publicly available dataset with 200 subjects.

Aims: Sepsis is a wide-spread severe problem faced in intensive care units (ICUs). Patients with undetected infections develop a complex reaction to fight off pathogens, which often cause organ damage and failure, leading to a life-threatening condition and death in many cases. The difficulty lies in early-detection as the infection does not present obvious symptoms. In this work we aim at assessing machine learning (ML) methods for early prediction of sepsis using electronic patient records (EPRs) provided by the Physionet Computing in Cardiology Challenge 2019.

Current sleep medicine relies on the supervised analysis of polysomnographic recordings, which comprise amongst others electroencephalogram (EEG), electromyogram (EMG), and electrooculogram (EOG) signals. Convolutional neural networks (CNN) provide an interesting framework for automated sleep classification, however, the lack of interpretability of its results has hampered CNN's further use in medicine. In this study, we train a CNN using as input Continuous Wavelet transformed EEG, EOG and EMG recordings from a publicly available dataset. The network achieved a 10-fold cross-validation Cohen's Kappa score of $\kappa = 0.71 \pm 0.01$. Further, we provide insights on how this network classifies individual epochs of sleep using Guided Gradient-weighted Class Activation Maps (Guided Grad-CAM). The proposed approach is able to produce fine-grained activation maps on time-frequency domain for each signal providing a useful tool for identifying relevant features in CNNs.

—This study aims to develop automated diagnostic tools to aid in the identification of rapid-eye-movement (REM) sleep behaviour disorder (RBD). Those diagnosed with RBD enact their dreams and therefore provide a unique characteristic of movement during REM sleep. To evaluate the characteristics of the disorder publicly available PSG recordings were used in combination with additional recordings sourced from multiple institutions. This combination of PSG recordings includes 44 healthy controls and 44 RBD subjects. This study aims to develop a robust diagnostic tool to analyse single channel electromyogram (EMG) recordings and manually annotated sleep stages to distinguish healthy controls from RBD subjects. Numerous techniques exist that can objectively quantify movement from EMG signals and are able to identify RBD subjects, which are further validated in this study. Using these established techniques in combination with additional features that the relationship of muscle movement between sleep stages and the general sleep architecture, the performance of RBD identification was shown to improve. This collection of objective EMG features used with a random forest classifier achieves RBD detection with 92% accuracy, 93% sensitivity, and 91% specificity.

We propose in this work a feature learning approach using deep bidirectional recurrent neural networks (RNNs) with attention mechanism for single-channel automatic sleep stage classification. We firstly decompose an EEG epoch into multiple small frames and subsequently transform them into a sequence of frame-wise feature vectors. Given the training sequences, the attention-based RNN is trained in a sequence-to-label fashion for sleep stage classification. Due to discriminative training, the network is expected to encode information of an input sequence into a high-level feature vector after the attention layer. We, therefore, treat the trained network as a feature extractor and extract these feature vectors for classification which is accomplished by a linear SVM classifier. We also propose a discriminative method to learn a filter bank with a DNN for preprocessing purpose. Filtering the frame-wise feature vectors with the learned filter bank beforehand leads to further improvement on the classification performance. The proposed approach demonstrates good performance on the Sleep-EDF dataset.

Current sleep medicine relies on the supervised analysis of polysomnographic measurements, comprising amongst others electroencephalogram (EEG), electromyogram (EMG), and electrooculogram (EOG) signals. Convolutional neural networks (CNN) provide an interesting framework to automated classification of sleep based on these raw waveforms. In this study, we compare existing CNN approaches to four databases of pathological and physiological subjects. The best performing model resulted in Cohen's Kappa of $\kappa = 0 .75$ on healthy subjects and $\kappa = 0 .64$ on patients suffering from a variety of sleep disorders. Further, we show the advantages of additional sensor data (i.e., EOG and EMG). Deep learning approaches require a lot of data which is scarce for less prevalent diseases. For this, we propose a transfer learning procedure by pretraining a model on large public data and fine-tune this on each subject from a smaller dataset. This procedure is demonstrated using a private REM Behaviour Disorder database, improving sleep classification by 24.4%.

Correctly identifying sleep stages is important in diagnosing and treating sleep disorders. This paper proposes a joint classification-and-prediction framework based on convolutional neural networks (CNNs) for automatic sleep staging, and, subsequently, introduces a simple yet efficient CNN architecture to power the framework. Given a single input epoch, the novel framework jointly determines its label (classification) and its neighboring epochs’ labels (prediction) in the contextual output. While the proposed framework is orthogonal to the widely adopted classification schemes, which take one or multiple epochs as contextual inputs and produce a single classification decision on the target epoch, we demonstrate its advantages in several ways. First, it leverages the dependency among consecutive sleep epochs while surpassing the problems experienced with the common classification schemes. Second, even with a single model, the framework has the capacity to produce multiple decisions, which are essential in obtaining a good performance as in ensemble-of-models methods, with very little induced computational overhead. Probabilistic aggregation techniques are then proposed to leverage the availability of multiple decisions. To illustrate the efficacy of the proposed framework, we conducted experiments on two public datasets: Sleep-EDF Expanded (Sleep-EDF), which consists of 20 subjects, and Montreal Archive of Sleep Studies (MASS) dataset, which consists of 200 subjects. The proposed framework yields an overall classification accuracy of 82.3% and 83.6%, respectively. We also show that the proposed framework not only is superior to the baselines based on the common classification schemes but also outperforms existing deep-learning approaches. To our knowledge, this is the first work going beyond the standard single-output classification to consider multitask neural networks for automatic sleep staging. This framework provides avenues for further studies of different neural-network architectures for automatic sleep staging.

As the collection of mobile health data becomes pervasive, missing data can make large portions of datasets inaccessible for analysis. Missing data has shown particularly problematic for remotely diagnosing and monitoring Parkinson's disease (PD) using smartphones. This contribution presents multi-source ensemble learning, a methodology which combines dataset deconstruction with ensemble learning and enables participants with incomplete data (i.e., where not all sensor data is available) to be included in the training of machine learning models and achieves a 100% participant retention rate. We demonstrate the proposed method on a cohort of 1513 participants, 91.2% of which contributed incomplete data in tapping, gait, voice, and/or memory tests. The use of multi-source ensemble learning, alongside convolutional neural networks (CNNs) capitalizing on the amount of available data, increases PD classification accuracy from 73.1% to 82.0% as compared to traditional techniques. The increase in accuracy is found to be partly caused by the use of multi-channel CNNs and partly caused by developing models using the large cohort of participants. Furthermore, through bootstrap sampling we reveal that feature selection is better performed on a large cohort of participants with incomplete data than on a small number of participants with complete data. The proposed method is applicable to a wide range of wearable/remote monitoring datasets that suffer from missing data and contributes to improving the ability to remotely monitor PD via revealing novel methods of accounting for symptom heterogeneity.

We present in this paper an efficient convolutional neural network (CNN) running on time-frequency image features for automatic sleep stage classification.Opposing to deep architectures which have been used for the task, the proposed CNN is much simpler.However, the CNN's convolutional layer is able to support convolutional kernels with different sizes, and therefore, capable of learning features at multiple temporal resolutions.In addition, the 1-max pooling strategy is employed at the pooling layer to better capture the shift-invariance property of EEG signals.We further propose a method to discriminatively learn a frequency-domain filter bank with a deep neural network (DNN) to preprocess the time-frequency image features.Our experiments show that the proposed 1-max pooling CNN performs comparably with the very deep CNNs in the literature on the Sleep-EDF dataset.Preprocessing the time-frequency image features with the learned filter bank before presenting them to the CNN leads to significant improvements on the classification accuracy, setting the stateof-the-art performance on the dataset.

Aims: Sleep arousals are typically defined as short intrusions of wakefulness into sleep that are presently quantified by manual sleep staging of electroencephalograms (EEGs) by trained clinicians. Due to mounting evidence that relates arousals to sleep disorders, there has been a growing interest in automating this approach. In this study, we aim to detect non-apnea events, respiratory-effort-related-arousals (RERA), and non-RERA arousals from a range of physiological signals recorded during a single night of sleep for the Physionet/Computing in Cardiology Challenge 2018. Methods: The proposed approach made use of the following signals: oxygen saturation (SaO2), electrocardiogram (ECG), averaged central EEG, electrooculogram (EOG) and electromyogram. After a simple bandpass/notch preprocessing using FIR filters, each raw input was segmented into 10 s epochs with 1 s shifts. 145 features were calculated for each epoch such as heart rate variability metrics (ECG) and spectral band powers (EEG/EOG). Dimension reduction was then performed using the first 20 components of Principal Component Analysis (PCA), and a Hidden Markov Model (HMM) trained on each subject. A 10-fold cross-validation was used to assess the method’s performance using sensitivity (SE), positive predictive value (PPV), area under curve (AUC) and F1-score (F1). For each fold, the trained HMM models were averaged and applied to the validation set. Segments with undefined labels were excluded from training. Results and Discussion: Table 1 shows the cross-validation results. While sensitivity was high, F1 remained at 11.2%. This suggests that the model has a propensity to classify epochs as arousals and reveals the need for a more informative set of features. A strong overlap in the features between arousals and non-arousals exists (even after PCA) as well as class imbalance with arousal periods occurring during <7% of time. Future work will focus on extending the current HMM approach and using deep learning techniques to identify such features in a more data-driven fashion, by better utilising the high dimensionality of the Challenge dataset.