Video-based actigraphy is a non-contact technology that measures body movement through video recordings, commonly expressed as activity count. It has been widely used to assess infant sleep. One validated method for video-based actigraphy is the three-dimensional recursive search (3DRS). However, its sleep-wake classification performance in preterm infants is only moderately satisfactory, necessitating methods to improve it. This paper proposes an enhanced approach to compute activity count for video-based actigraphy in evaluating sleep patterns in preterm infants. The proposed technique involves applying exponentially weighted moving average (EWMA) to continuous activity count to reduce abrupt variations and false detection of wakefulness. Additionally, the 3DRS protocol we previously utilized is proprietary, limiting its accessibility to the public and its widespread adoption. To overcome this limitation, we have investigated the efficacy of an open-source algorithm known as background subtraction (BS), which, in our previous work, has demonstrated superior performance in body motion detection in preterm infants. To evaluate the performance of the proposed method, a dataset consisting of video recordings from five hospitalized preterm infants was used. The activity counts obtained from 3DRS, and BS were compared in terms of their ability to classify sleep and wake states using a linear discriminant classifier. The results obtained through leave-one-patient-out cross-validation revealed a significant improvement in the classification of sleep and wake states in preterm infants for both methods when the EWMA were applied to the activity counts. The mean Cohen's kappa coefficients were found to be 0.58 for 3DRS and 0.51 for BS. Despite BS exhibited comparatively lower performance than 3DRS, it can still be considered a viable alternative for acquiring video-based actigraphy data in the assessment of sleep of preterm infants.

In this paper, we concentrate on the per-class accuracy of neural network-based classification in the context of identifying acoustic environments. Even a fully supervised learning framework with an equal amount of data for each class can lead to significant differences in class accuracies. This is then amplified by semi-supervised learning using naturally imbalanced data. To address this problem, we propose an adaptive method for pseudo-label selection via a straightforward optimization of the validation accuracy per class, aimed specifically at reducing the variance between different classes. The proposed method is general and can be applied for both maximum probability and entropy-based confidence criteria. Compared to fully supervised learning as well as state-of-the-art methods for pseudo-labeling, it achieves the lowest variances of per-class accuracy and the highest accuracies of the minority classes when tested on common publicly available environment sound databases.

OBJECTIVE: Photoplethysmography (PPG) enables unobtrusive heart rate monitoring, which can be used in wrist-worn applications. Its potential for detecting atrial fibrillation (AF) has been recently presented. Besides AF, another cardiac arrhythmia increasing stroke risk and requiring treatment is atrial flutter (AFL). Currently, the knowledge about AFL detection with PPG is limited. The objective of our study was to develop a model that classifies AF, AFL, and sinus rhythm with or without premature beats from PPG and acceleration data measured at the wrist in daily life. METHODS: A dataset of 40 patients was collected by measuring PPG and accelerometer data, as well as electrocardiogram as a reference, during 24-hour monitoring. The dataset was split into 75%-25% for training and testing a Random Forest (RF) model, which combines features from PPG, inter-pulse intervals (IPI), and accelerometer data, to classify AF, AFL, and other rhythms. The performance was compared to an AF detection algorithm combining traditional IPI features for determining the robustness of the accuracy in presence of AFL. RESULTS: The RF model classified AF/AFL/other with sensitivity and specificity of 97.6/84.5/98.1% and 98.2/99.7/92.8%, respectively. The results with the IPI-based AF classifier showed that the majority of false detections were caused by AFL. CONCLUSION: The PPG signal contains information to classify AFL in the presence of AF, sinus rhythm, or sinus rhythm with premature contractions. SIGNIFICANCE: PPG could indicate presence of AFL, not only AF.

This paper presents an actigraphy-based approach for sleep/wake detection for insomniacs. Due to its relative unobtrusiveness, actigraphy is often used to estimate overnight sleep-wake patterns in clinical practice. However, its performance has been shown to be limited in subjects with sleep complaints such as insomniacs. Quantifying activity counts on 30-s epoch basis, as usually done in regular actigraphy, may lead to an underestimation of wake periods where the subject shows reduced body movements. We therefore propose a new actigraphic feature to characterize the `possibility' of epochs being asleep (or awake) before or after its nearest epoch with a very high activity levels. It is expected to correctly identify some wake epochs when they are very close to the high activity epochs, although they can be motionless. A data set containing 25 insomnia subjects and a linear discriminant classifier were used to test our approach in this study. Leave-one-subject-out cross validation results show that combining the new and the traditional actigraphic features led to a markedly improved performance in sleep/wake detection compared to that using the traditional feature only, with an increase in Cohen's kappa from 0.49 to 0.55.

This paper explores the probabilistic properties of sleep stage sequences and transitions to improve the performance of sleep stage detection using cardiorespiratory features. A new classifier, based on conditional random fields, is used in different sleep stage detection tasks (N3, NREM, REM, and wake) in night-time recordings of electrocardiogram and respiratory inductance plethysmography of healthy subjects. Using a dataset of 342 polysomnographic recordings of healthy subjects, among which 135 with regular sleep architecture, it outperforms hidden Markov models and Bayesian linear discriminants in all tasks, achieving an average accuracy of 87.38% and kappa of 0.41 (87.27% and 0.49 for regular subjects) for N3 detection, 78.71% and 0.55 (80.34% and 0.56 for regular subjects) for NREM detection, 88.49% and 0.51 (87.35% and 0.57 for regular subjects) for REM, and 85.69% and 0.51 (90.42% and 0.52 for regular subjects) for wake. In comparison with the state of the art, and having been tested on a much larger dataset, the classifier was found to outperform most of the work reported in the literature for some of the tasks, in particular for subjects with regular sleep architecture. It achieves a comparable accuracy for N3, higher accuracy and kappa for REM, and higher accuracy and comparable kappa for NREM than the best performing classifiers described in the literature.

Human slow wave sleep (SWS) during bedtime is paramount for energy conservation and memory consolidation. This study aims at automatically detecting SWS from nocturnal sleep using cardiorespiratory signals that can be acquired with unobtrusive sensors in a home-based scenario. From the signals, time-dependent features are extracted for continuous 30-s epochs. To reduce the measuring noise, body motion artifacts, and/or within-subject variability in physiology conveyed by the features, and thus, enhance the detection performance, we propose to smooth the features over each night using a spline fitting method. In addition, it was found that the changes in cardiorespiratory activity precede the transitions between SWS and the other sleep stages (non-SWS). To this matter, a novel scheme is proposed that performs the SWS detection for each epoch using the feature values prior to that epoch. Experiments were conducted with a large dataset of 325 overnight polysomnography (PSG) recordings using a linear discriminant classifier and tenfold cross validation. Features were selected with a correlation-based method. Results show that the performance in classifying SWS and non-SWS can be significantly improved when smoothing the features and using the preceding feature values of 5-min earlier. We achieved a Cohen's Kappa coefficient of 0.57 (at an accuracy of 88.8%) using only six selected features for 257 recordings with a minimum of 30-min overnight SWS that were considered representative of their habitual sleeping pattern at home. These features included the standard deviation, low-frequency spectral power, and detrended fluctuation of heartbeat intervals as well as the variations of respiratory frequency and upper and lower respiratory envelopes. A marked drop in Kappa to 0.21 was observed for the other nights with SWS time of less than 30 min, which were found to more likely occur in elderly. This will be the future challenge in cardiorespiratory-based SWS detection.

This paper proposes the use of dynamic warping (DW) methods for improving automatic sleep and wake classification using actigraphy and respiratory effort. DW is an algorithm that finds an optimal nonlinear alignment between two series allowing scaling and shifting. It is widely used to quantify (dis)similarity between two series. To compare the respiratory effort between sleep and wake states by means of (dis)similarity, we constructed two novel features based on DW. For a given epoch of a respiratory effort recording, the features search for the optimally aligned epoch within the same recording in time and frequency domain. This is expected to yield a high (or low) similarity score when this epoch is sleep (or wake). Since the comparison occurs throughout the entire-night recording of a subject, it may reduce the effects of within- and between-subject variations of the respiratory effort, and thus help discriminate between sleep and wake states. The DW-based features were evaluated using a linear discriminant classifier on a dataset of 15 healthy subjects. Results show that the DW-based features can provide a Cohen's Kappa coefficient of agreement κ = 0.59 which is significantly higher than the existing respiratory-based features and is comparable to actigraphy. After combining the actigraphy and the DW-based features, the classifier achieved a κ of 0.66 and an overall accuracy of 95.7%, outperforming an earlier actigraphy- and respiratory-based feature set ( κ = 0.62). The results are also comparable with those obtained using an actigraphy- and cardiorespiratory-based feature set but have the important advantage that they do not require an ECG signal to be recorded.

This paper describes a method to adapt the spectral features extracted from heart rate variability (HRV) for sleep and wake classification. HRV series can be derived from electrocardiogram (ECG) signals obtained from single-night polysomnography (PSG) recordings. Traditionally, the HRV spectral features are extracted from the spectrum of an HRV series with fixed boundaries specifying bands of very low frequency (VLF), low frequency (LF), and high frequency (HF). However, because they are fixed, they may fail to accurately reflect certain aspects of autonomic nervous activity, which in turn may limit their discriminative power when using HRV spectral features, e.g., in sleep and wake classification. This is in part related to the fact that the sympathetic tone (partially reflected in the LF band) and the respiratory activity (modulated in the HF band) will vary over time. In order to minimize the impact of these differences, we adapt the HRV spectral boundaries using time-frequency analysis. Experiments conducted on a dataset acquired from 15 healthy subjects show that the discriminative power of the adapted HRV spectral features are significantly increased when classifying sleep and wake. Additionally, this method also provides a significant improvement of the overall classification performance when used in combination with some other (non-spectral) HRV features.

Four time-frequency and time-scale methods are studied for their ability of detecting myoclonic seizures from accelerometric data. Methods that are used are: the short-time Fourier transform (STFT), the Wigner distribution (WD), the continuous wavelet transform (CWT) using a Daubechies wavelet, and a newly introduced model-based matched wavelet transform (MOD). Real patient data are analyzed using these four time-frequency and time-scale methods. To obtain quantitative results, all four methods are evaluated in a linear classification setup. Data from 15 patients are used for training and data from 21 patients for testing. Using features based on the CWT and MOD, the success rate of the classifier was 80%. Using STFT or WD-based features, the classification success is reduced. Analysis of the false positives revealed that they were either clonic seizures, the onset of tonic seizures, or sharp peaks in "normal" movements indicating that the patient was making a jerky movement. All these movements are considered clinically important to detect. Thus, the results show that both CWT and MOD are useful for the detection of myoclonic seizures. On top of that, MOD has the advantage that it consists of parameters that are related to seizure duration and intensity that are physiologically meaningful. Furthermore, in future work, the model can also be useful for the detection of other motor seizure types.

Context awareness will become increasingly important in future domestic consumer electronics. In many domestic context aware applications, there is a need for location and orientation information about persons, devices or objects in the home. This information can be provided by a dedicated indoor location system. A consumer location system should be robust, safe, easy to set up, low cost, and should have a minimal infrastructure. In this paper, a step towards consumer location systems is taken by proposing an ultrasonic positioning method that needs just a single compact base station to measure 3D positions of mobile devices in a room. The method employs ultrasound time-of-flight tri-lateration to estimate device positions, using a base station containing an array of three ultrasound transducers. Five potential problems of the proposed method are identified; the main problem being line-of-sight path occlusion. The method has been prototyped, and initial results show an accuracy of 1.41 m or better for 95% of the position estimates, in case of a good line-of-sight path. It is concluded that the method is promising for providing 3D position information at around 1 m accuracy for context aware applications, but that the problem of line-of-sight occlusion requires further investigation.

In many (audio) processing algorithms, involving manipulation of discrete-time signals, the performance can vary strongly over the repertoire that is used. This may be the case when the signals from the various channels are allowed to be strongly positively or negatively correlated. We propose and analyze a general formula for tracking the (time-dependent) correlation between two signals. Some special cases of this formula lead to classical results known from the literature, others are new. This formula is recursive in nature, and uses only the instantaneous values of the two signals, in a low-cost and low-complexity manner; in particular, there is no need to take square roots or to carry out divisions. Furthermore, this formula can be modified with respect to the occurrence of the two signals so as to further decrease the complexity, and increase ease of implementation. The latter modification comes at the expense that not the actual correlation is tracked, but, rather, a somewhat deformed version of it. To overcome this problem, we propose, for a number of instances of the tracking formula, a simple warping operation on the deformed correlation. Now we obtain, at least for sinusoidal signals, the correct value of the correlation coefficient. Special attention is paid to the convergence behavior of the algorithm for stationary signals and the dynamic behavior if there is a transition to another stationary state; the latter is considered to be important to study the tracking abilities to nonstationary signals. We illustrate tracking algorithm by using it for stereo music fragments, obtained from a number of digital audio recordings.