Wearable sensing devices, such as Holter monitors, will play a crucial role in the future of digital health. Unsupervised learning frameworks such as Self-Supervised Learning (SSL) are essential to map these single-lead electrocardiogram (ECG) signals with their anticipated clinical outcomes. These signals are characterized by a tempo-variant component whose patterns evolve through the recording and an invariant component with patterns that remain unchanged. However, existing SSL methods only drive the model to encode the invariant attributes, leading the model to neglect tempo-variant information which reflects subject-state changes through time. In this paper, we present Parallel-Learning of Invariant and Tempo-variant Attributes (PLITA), a novel SSL method designed for capturing both invariant and tempo-variant ECG attributes. The latter are captured by mandating closer representations in space for closer inputs on time. We evaluate both the capability of the method to learn the attributes of these two distinct kinds, as well as PLITA ’s performance compared to existing SSL methods for ECG analysis. PLITA performs significantly better in the set-ups where tempo-variant attributes play a major role.

We present EarEOG, a real-time wheelchair control system using around-ear electrooculogram (EOG) signals. Electrodes are placed in standard over-the-ear headphones to improve user comfort. By detecting around-the-ear signals from eye gestures and jaw clenching, EarEOG offers a non-invasive and intuitive approach to low-latency wheelchair control. We describe the methods for signal acquisition, as well as the algorithms used for signal processing and classification. The feasibility, robustness, and low latency of EarEOG were confirmed through two experiments. The algorithm demonstrated a classification accuracy of 94.1% for all motion signals, which further improved to 97.3% when personalized models were applied. To ensure stability, we examined electrode impedance and algorithm accuracy across multiple trials where participants operated simulated wheelchairs while wearing EarEOG. The results indicated that when the electrode impedance was below 1 MΩ, all participants successfully controlled the simulated wheelchair. Furthermore, EarEOG demonstrated low latency, with recognition delays of less than 125 ms.

To investigate the contextual and temporal distribution of false positives (FPs) in a state-of-the-art deep learning (DL)-based atrial fibrillation (AF) detection algorithm when applied to an electrocardiogram (ECG) dataset collected under free-living ambulatory conditions. We hypothesize that under such conditions, the FPs detected by a DL model might have some correlations with the patient’s ambulatory contexts. First, a DL model is trained and evaluated on three public arrhythmia datasets from PhysioNet. It is ensured that the model has state-of-the-art performance on these public datasets. Thereafter, the same model is applied to a 215-days long contextualized single-channel ECG dataset collected under free-living ambulatory conditions. Through a manual examination of the model’s output, ground truth is obtained and the correlations between the patient’s ambulatory contexts and the true/false positive rate are analyzed. Nearly 62% of the segments marked as AF by the model were ≤50 seconds in length, and 99.9% of them were FPs. Among these non-trivial short segments of FPs, almost 78% were mainly associated with three specific contextual events; change in activity, change in body position (especially during the night), and sudden movement acceleration. Moreover, the number of FPs detected by the DL model are higher in female than in male participants. Finally, true positive (TP) AF segments are found more in the morning and late evening. These findings may have significant implications for the current use and future design of DL models for AF detection, and help understand the role of context information in reducing the FP rate in real-time AF detection under free-living conditions.

A dry-electrode head-mounted sensor for visually-evoked electroencephalogram (EEG) signals has been introduced to the gamer market, and provides wireless, low-cost tracking of a user’s gaze fixation on target areas in real-time. Unlike traditional EEG sensors, this new device is easy to set up for non-professionals. We conducted a Fitts’ law study (N = 6) and found the mean throughput (TP) to be 0.82 bits/s. The sensor yielded robust performance with error rates below 1%. The overall median activation time (AT) was 2.35 s with a minuscule difference between one or nine concurrent targets. We discuss whether the method might supplement camera-based gaze interaction, for example, in gaze typing or wheelchair control, and note some limitations, such as a slow AT, the difficulty of calibration with thick hair, and the limit of 10 concurrent targets.

Classification of non-stationary electroencephalogram (EEG) data are of utmost importance for brain-computer interface (BCI) technology. This paper proposes a robust multiclass motor imagery (MI) BCI data classification technique. It is based on filter bank common spatial patterns (FBCSP) and AdaBoost classification technique. The method is tested on the 4-class MI BCI competition IV dataset 2a and the results show superior performance compared to the current state-of-the-art performances. This paper also analyzes different frequency sub-bands for the MI EEG data, in order to find the best sub-band which contains the most significant features for distinguishing different MI tasks.

Detection of P-waves in electrocardiogram (ECG) signals is of great importance to cardiologists in order to help them diagnosing arrhythmias such as atrial fibrillation. This paper proposes an end-to-end deep learning approach for detection of P-waves in ECG signals. Four different deep Recurrent Neural Networks (RNNs), namely, the Long-Short Term Memory (LSTM) are used in an ensemble framework. Each of these networks are trained to extract the useful features from raw ECG signals and determine the absence/presence of P-waves. Outputs of these classifiers are then combined for final detection of the P-waves. The proposed algorithm was trained and validated on a database which consists of more than 111000 annotated heart beats and the results show consistently high classification accuracy and sensitivity of around 98.48% and 97.22%, respectively.

An efficient implementation of a multi-class motor imagery (MI) brain computer interface (BCI) classification scheme is presented in this work. The proposed method uses the common spatial pattern (CSP) and filter bank CSP (FBCSP) algorithms, with both one versus all (OVA) and one versus one (OVO) approach for multi-class extension. Mutual information (MInf) based feature selection algorithm has been used to obtain the features to train different linear discriminant analysis (LDA) classifiers. To improve the performance, the outputs of these classifiers are combined using two statistical methods: the mode of the OVA and OVO classifiers, and the more sophisticated Dempster-Shafer (DS) theory. The method has been evaluated on the 4-class MI dataset (BCI competition IV 2a), and the results showed that it has outperformed the winner of the competition (maximum kappa value of 0.593 vs 0.569). The proposed method proved the benefits of combining classifiers with appropriate techniques.

Objectives: Electrocardiogram (ECG) is an important clinical tool for diagnosis of cardiac abnormalities. Physicians make diagnoses by visual examination of ECGs. Analysing huge amounts of ECGs however, can be very time consuming and cumbersome. Hence, developing analytic software is of great importance to automatically analyse these ECG signals to detect efficiently the common cardiac arrhythmias. Methods: Proposed an ensemble learning approach for automatic processing of ECG signals and classification of arrhythmias. Twenty six features (based on wavelets, heartbeat intervals, and RR-intervals) are extracted and three algorithms, namely, Random Forest (RF), Adaptive Boosting (AdaBoost) and Artificial Neural Network (ANN) are utilized for classification. Results: The proposed method is evaluated on ECG signals from 44 recordings of the MIT-BIH arrhythmia database. The overall classification accuracy of the RF, AdaBoost, and ANN are 96.16%, 96.16% and 94.49%, respectively. Additionally, the overall classification accuracy of the ensemble model is improved to 96.18%. Conclusion: Experimental results show that the performance of the ensemble model for ECG heartbeat classification improves the overall accuracy. Significance: This paper proposes an accurate and easy to use approach to classify heartbeats into one of the five classes recommended by ANSI/AAMI standard, which can be used in real-time within a tele-health monitoring framework.

Frontal cortical activity can be used as an indicator of mental stress. Using the frontal alpha activity in the electroencephalogram (EEG) as a basis, the Frontal Brain Asymmetry, Resting Brain Asymmetry, Frontal Asymmetry Index and Alpha Band Power along with Heart Rate Variability have been used as features in the present work to analyse the level of stress in subjects. Eleven healthy subjects participated in the experiments with 3 tests each based on 4 trials, with each trial posing an added cognitive challenge. Trials consisted of the Hopkins Verbal Learning Test, Trial Making Test, and Corsi Block Tapping Test. They were tested on factors based on verbal understanding, memory, attention, focus and overall cognitive performance. Using a rule-based classification model and combining the results from the features, 3 categories (low, medium and high) of stresses were identified. Partial consistency through the various extracted features indicate that subjects vary in their ability to manage stress. Overall, 2 subjects were indicative of a low stressed state, 6 subjects showed medium stress levels and 3 subjects were categorized under the high stress state.

Brain Computer Interface (BCI) systems enable subjects affected by neuromuscular disorders to interact with the outside world. A P300 speller uses Event Related Potential (ERP) components, generated in the brain in the presence of a target stimulus, to extract information about the user's intent. Several methods have been proposed for spatial filtering and classification of the P300 components. In this study, xDAWN algorithm, Independent Component Analysis (ICA) and Principal Component Analysis (PCA) methods are used and evaluated based on the classification performance of two different classifiers, namely the Support Vector Machine (SVM) and Fisher's Linear Discriminant Analysis (FLDA). In addition, it is shown that the incorporation of some prior knowledge regarding the location of P300 elicitation on the scalp can reduce the computational load while maintaining or even improving the classification performance.

In this paper, we propose a semi-blind channel estimation algorithm for amplify-and-forward (AF) relay networks. The algorithm fits well for the recently developed space-time coding (STC) technique in AF relay network that serves for small size terminal and achieve the transmission diversity. Compared to the optimal training based estimators, e.g., maximum likelihood (ML) or linear minimum mean square (MMSE), the proposed semi-blind approach requires less training for successful channel estimation, or it yields better estimates if the same amount of training is used. The channel ambiguity issue as well as its relationship with the traditional semi-blind method is discussed in detail. We then provide various numerical examples to corroborate the proposed studies.

Functional Magnetic Resonance Imaging (fMRI) is an important technique to study the human brain (the most complex biological dynamical system) functions which are often described by the hemodynamic responses (HDR). It measures the changes of the blood oxygenation level dependent (BOLD) signals due to the neural activities. The measured fMRI data is the response of the human brain to a particular processing task. In this paper, the nonlinear autoregressive with exogenous inputs (NARX) neural networks are investigated as a method to model the dynamics underlying the fMRI data. Studies on both simulated as well as real event-related fMRI data show that the proposed scheme can capture the underlying dynamics of the brain and reconstruct the BOLD signals from the measured noisy fMRI data. In addition, a good estimate of the HDR of the brain is also obtained.

This paper presents spatio-temporal modeling and analysis methods to fMRI data. Based on the nonlinear autoregressive with exogenous inputs (NARX) model realized by the Bayesian radial basis function (RBF) neural networks, two methods (NARX-1 and NARX-2) are proposed to capture the unknown complex dynamics of the brain activities. Simulation results on both synthetic and real fMRI data clearly show that the proposed schemes outperform the conventional t-test method in detecting the activated regions of the brain.

This paper deals with the problem of multiuser detection in direct-sequence code-division multiple-access (DS-CDMA) systems in multipath environments. The existing multiuser detectors can be divided into two categories: (1) low-complexity poor-performance linear detectors and (2) high-complexity good-performance nonlinear detectors. In particular, in channels where the orthogonality of the code sequences is destroyed by multipath, detectors with linear complexity perform much worse than the nonlinear detectors. In this paper, we propose an enhanced multisurface method (EMSM) for multiuser detection in multipath channels. EMSM is an intermediate piecewise linear detection scheme with a run-time complexity linear in the number of users. Its bit error rate performance is compared with existing linear detectors, a nonlinear radial basis function detector trained by the new support vector learning algorithm, and Verdu's optimal detector. Simulations in multipath channels, for both synchronous and asynchronous cases, indicate that it always outperforms all other linear detectors, performing nearly as well as nonlinear detectors.

Two adaptive algorithms (time varying and exponentially weighted) based on the H/sup /spl infin// principles are proposed for the minimization of electrooculogram (EOG) artifacts from corrupted electroencephalographic signals. Performance of the proposed algorithms are compared with the least-mean-square (LMS) algorithm. Improvements in the output signal-to-noise ratio along with time plots are used for the comparison. It is found that the H/sup /spl infin//-based algorithms effectively minimize the EOG artifacts and always outperform the LMS algorithm.

In this paper, a novel direct sequence/spread spectrum (DS/SS) communication system is proposed. This scheme exploits the two-dimensional complex valued chaotic Ikeda map as the spreading sequences. With this double spreading DS/SS system, the effect of multiple access interference can be mitigated by choosing the spreading sequences with appropriate cross-correlation properties. The performance of the system is assessed and demonstrated in a multiuser environment by means of computer simulation with additive white Gaussian noise, Rayleigh fading, and selective fading channel conditions. These studies reveal that the proposed system significantly outperforms the Gold code DS/SS-BPSK system in synchronous channel conditions. In asynchronous case, the improvement is substantial for low signal-to-noise ratios.

In this paper, a novel quadrature modulation based complex digital processing (CDP) DS/SS system is proposed which exploits the 2-D complex chaotic Ikeda map (IM) as the spreading sequence. With the proposed system, we can mitigate multiple access interference (MAI) by using complex-valued processing method for the spreading and despreading operations. The performance of the proposed CDP system is evaluated in multiuser environment numerically and analytically under additive white Gaussian noise (AWGN) and Rayleigh flat fading channel conditions. Simulation results reveal that the proposed CDP DS/SS system with the IM spreading sequence significantly outperforms the conventional Gold code based inphase-quadrature (I-Q) DS/SS system.

Parallel interference cancellation (PIC) detector is designed to subtract the multi-access interference (MAI), and hence, it has the potential to achieve further performance improvement for direct-sequence code-division multiple access (DS-CDMA) systems. However, its performance depends on the accuracy of its data estimation. In the conventional PIC (CPIC) detector, the data estimation is based on the matched filter (MF), which results in poor performance. Moreover, in multi-cell environment, CPIC suffers from the interference from other cells. These disadvantages limit the potential of PIC. In this paper, a new PIC scheme is proposed and examined in detail, which uses blind adaptive minimum mean square error (MMSE) detectors instead of MFs for data estimation in each cancellation stage. Theoretical performance analyses as well as numerical simulations in varied situations reveal that the scheme has much improved performance over the CPIC. More importantly, it is immune to the near-far problem and is able to suppress the interference from other cells.

Multiuser detection is an effective method to suppress the multiple access interference (MAI) and improve the performance of direct sequence code division multiple access (DS-CDMA) communication systems. Since the optimum decision boundary is nonlinear structure in the optimum multiuser detector can improve the performance of the receiver. The authors of this paper have proposed two nonlinear blind adaptive interference cancellation algorithms (the exact Newton and the approximate Newton) in one of their earlier works [Guo Linlin and Sadasivan,P.K., Sept. 2002]. In this paper, a complete performance analysis and characterization of the nonlinear schemes with respect to various asynchronous parameters are investigated in both synchronous and asynchronous DS-CDMA systems. The numerical results show that the adaptive blind multi-user detectors yield significant bit-error-rate (BER) improvement over the conventional matched filter detector in the presence of strong MAI. Further, the two nonlinear adaptive algorithms outperform the linear algorithm due to the nonlinear nature of the transmitted signals. Most attractively, the approximate Newton algorithm offers lower computational complexity, higher numerical stability and similar BER performance in comparison to the exact Newton algorithm.