Ruhi Mahajan

Unsupervised Eye Blink Artifact Denoising of EEG Data with Modified Multiscale Sample Entropy, Kurtosis, and Wavelet-ICA

Brain activities commonly recorded using the electroencephalogram (EEG) are contaminated with ocular artifacts. These activities can be suppressed using a robust independent component analysis (ICA) tool, but its efficiency relies on manual intervention to accurately identify the independent artifactual components. In this paper, we present a new unsupervised, robust, and computationally fast statistical algorithm that uses modified multiscale sample entropy (mMSE) and Kurtosis to automatically identify the independent eye blink artifactual components, and subsequently denoise these components using biorthogonal wavelet decomposition. A 95% two-sided confidence interval of the mean is used to determine the threshold for Kurtosis and mMSE to identify the blink related components in the ICA decomposed data. The algorithm preserves the persistent neural activity in the independent components and removes only the artifactual activity. Results have shown improved performance in the reconstructed EEG signals using the proposed unsupervised algorithm in terms of mutual information, correlation coefficient, and spectral coherence in comparison with conventional zeroing-ICA and wavelet enhanced ICA artifact removal techniques. The algorithm achieves an average sensitivity of 90% and an average specificity of 98%, with average execution time for the datasets (N = 7) of 0.06 s (SD = 0.021) compared to the conventional wICA requiring 0.1078 s (SD = 0.004). The proposed algorithm neither requires manual identification for artifactual components nor additional electrooculographic channel. The algorithm was tested for 12 channels, but might be useful for dense EEG systems.

Comparative Study of Wavelet-Based Unsupervised Ocular Artifact Removal Techniques for Single-Channel EEG Data

Electroencephalogram (EEG) is a technique for recording the asynchronous activation of neuronal firing inside the brain with non-invasive scalp electrodes. Artifacts, such as eye blink activities, can corrupt these neuronal signals. While ocular artifact (OA) removal is well investigated for multiple channel EEG systems, in alignment with the recent momentum toward minimalistic EEG systems for use in natural environments, we investigate unsupervised and effective removal of OA from single-channel streaming raw EEG data. In this paper, the unsupervised wavelet transform (WT) decomposition technique was systematically evaluated for the effectiveness of OA removal for a single-channel EEG system. A set of seven raw EEG data set was analyzed. Two commonly used WT methods, Discrete Wavelet Transform (DWT) and Stationary Wavelet Transform (SWT), were applied. Four WT basis functions, namely, haar, coif3, sym3, and bior4.4, were considered for OA removal with universal threshold and statistical threshold (ST). To quantify OA removal efficacy from single-channel EEG, five performance metrics were utilized: correlation coefficients, mutual information, signal-to-artifact ratio, normalized mean square error, and time-frequency analysis. The temporal and spectral analysis shows that the optimal combination could be DWT with ST with coif3 or bior4.4 to remove OA among 16 combinations. This paper demonstrates that the WT can be an effective tool for unsupervised OA removal from single-channel EEG data for real-time applications.

Comparative analysis of wavelet based approaches for reliable removal of ocular artifacts from single channel EEG

For biomedical and scientific fields, Electroencephalography (EEG) has turned out to be an important tool to understand, study, and utilize brain functionalities. To fully utilize EEG signals in real-life closed-loop applications, artifacts such as ocular must be removed. Wavelet transform is one of the powerful methods to remove ocular artifacts from single channel EEG devices. In this study, both stationary and discrete wavelet transforms (SWT and DWT, respectively) have been compared with various wavelet basis functions, such as sym3, haar, coif3, and bior4.4 using either universal threshold (UT) or statistical threshold (ST). Different combinations of wavelet transform techniques, mother wavelets, and thresholds are compared to identify an optimum combination for ocular artifact removal. Performance metrics like Correlation Coefficient (CC), Normalized Mean Square Error (NMSE), Time Frequency Analysis, and execution time have been calculated for measuring the effectiveness of each combination. According to CC, DWT+UT combination turned out to be a good option for the ocular artifact removal. However, according to NMSE and time frequency analysis, SWT+ST has generated better performance in keeping neural segments of EEG unaffected. According to the measurement of execution times, DWT+ST is faster compared to other combinations. The study shows that wavelet transform is suitable in artifact removal from single channel EEG data to implement in ambulatory real-time EEG systems.

Real-time hybrid ocular artifact detection and removal for single channel EEG

Electroencephalography (EEG) is a promising technique to record brain activities in natural settings. However, EEG signals are usually contaminated by Ocular Artifacts (OA) such as eye blink activities. Removal of OA is critical to obtain clean EEG signals required for the feature extraction and classification. With the increasing interest in wearable technologies, single channel EEG systems are becoming more prevalent. Such ambulatory devices require real-time signal processing for immediate feedback. This paper presents a hybrid algorithm to detect and remove OA from single channel EEG signal using NeuroMonitor hardware platform. The algorithm first detects the eye blinks (OA zone) using Algebraic approach, and then removes artifact from OA zone using Discrete Wavelet Transform (DWT) decomposition method. De-noising technique is applied only to the OA zone to keep the critical neural information intact. The OA removal algorithm is applied to the online data for 0.5 sec epoch length. The performance evaluation is carried out qualitatively and quantitatively using time-frequency analysis, mean square coherence and other statistical parameters, i.e. Correlation Coefficient and Mutual Information. Processing time for DWT was significantly lower (x25) to that of SWT. This proposed hybrid OA removal algorithm demonstrates real-time execution with sufficient accuracy.

NeuroMonitor ambulatory EEG device: Comparative analysis and its application for cognitive load assessment

We have previously presented a wireless ambulatory EEG device (NeuroMonitor) to non-invasively monitor prefrontal cortex scalp EEG activity in real-life settings. This paper discusses analysis and application of data acquired using this device. We assess the device data against a commercially available, clinical grade Neuroscan SynAmps RT EEG system. For the comparison, temporal statistical measures and Power Spectral Density (PSD) are computed for the simultaneous recordings from both devices from (nearly) identical electrode locations. Although the analog signal processing, sampling, and data recording specifications are slightly different for these devices (e.g., filter specifications, ADC - NeuroMonitor: 16 bit and Neuroscan: 24 bit, electrodes - NeuroMonitor: GS26 Pre-gelled Disposable, Neuroscan: Ag/AgCl reusable EEG disc electrodes), the temporal signals and the PSD of two devices had sufficient correlation. The paper also describes pilot data collection for a test protocol to determine cognitive load using the NeuroMonitor device. For analyzing attention levels for 5 different tasks, EEG rhythms (Alpha, Beta and Theta) are extracted and cognitive load index (CLI) is computed. Results show variations in the PSD of these rhythms with respect to corresponding expected cognitive loads in attention-related and relaxed tasks. This study validates the NeuroMonitor ambulatory EEG device data and shows a use-case for real-life cognitive load studies.

Ambulatory EEG NeuroMonitor platform for engagement studies of children with development delays

Engagement monitoring is crucial in many clinical and therapy applications such as early learning preschool classes for children with developmental delays including autism spectrum disorder (ASD), attention-deficit hyperactivity disorder (ADHD), or cerebral palsy; as it is challenging for the instructors to evaluate the individual responses of these children to determine the effectiveness of the teaching strategies due to the diverse and unique need of each child who might have difficulty in verbal or behavioral communication. This paper presents an ambulatory scalp electroencephalogram (EEG) NeuroMonitor platform to study brain engagement activities in natural settings. The developed platform is miniature (size: 2.2” x 0.8” x 0.36”, weight: 41.8 gm with 800 mAh Li-ion battery and 3 snap leads) and low-power (active mode: 32 mA low power mode: under 5mA) with 2 channels (Fp1, Fp2) to record prefrontal cortex activities of the subject in natural settings while concealed within a headband. The signals from the electrodes are amplified with a low-power instrumentation amplifier; notch filtered (fc = 60Hz), then band-passed by a 2nd-order Chebyshev-I low-pass filter cascaded with a 2nd-order low-pass (fc = 125Hz). A PSoC ADC (16-bit, 256 sps) samples this filtered signal, and can either transmit it through a Class-2 Bluetooth transceiver to a remote station for real-time analysis or store it in a microSD card for offline processing. This platform is currently being evaluated to capture data in the classroom settings for engagement monitoring of children, aimed to study the effectiveness of various teaching strategies that will allow the development of personalized classroom curriculum for children with developmental delays.

Sample Entropy enhanced wavelet-ICA denoising technique for eye blink artifact removal from scalp EEG dataset

Scalp Electroencephalogram (EEG) recordings are usually contaminated with a variety of artifacts, which can be removed by threshold-based classifiers, Principal Component Analysis, Independent Component Analysis (ICA), wavelet-based multi-resolution analysis, or higher order statistics. In this paper we propose Sample Entropy, a self-sufficient statistical measure to identify the eye blink related independent components and Haar wavelet decomposition to subsequently denoise these components. The proposed method identified the blink artifactual components with an accuracy of 88% in our pilot study (N=4). The results demonstrated the improved performance of eye-blink artifacts removal with the neural activity intact in terms of Mutual Information (1.27 / 0.318 / 1.15), Correlation coefficient (0.574 / 0.369/ 0.569), and Standard deviation ratio (0.559/ 0.375 / 0.551) in comparison to standard Zeroing-ICA and wavelet-ICA based techniques, respectively. Instead of human expertise intervention to identify the eye blink components after Extended Infomax ICA decomposition, the algorithm offers potential for automation. This algorithm also offers advantage of being computationally fast and inexpensive, and does not require additional Electrooculographic signals for referencing.

Design and validation of a wearable “DRL-less” EEG using a novel fully-reconfigurable architecture

The conventional EEG system consists of a driven-right-leg (DRL) circuit, which prohibits modularization of the system. We propose a Lego-like connectable fully reconfigurable architecture of wearable EEG that can be easily customized and deployed at naturalistic settings for collecting neurological data. We have designed a novel Analog Front End (AFE) that eliminates the need for DRL while maintaining a comparable signal quality of EEG. We have prototyped this AFE for a single channel EEG, referred to as Smart Sensing Node (SSN), that senses brain signals and sends it to a Command Control Node (CCN) via an I2C bus. The AFE of each SSN (referential-montage) consists of an off-the-shelf instrumentation amplifier (gain=26), an active notch filter fc = 60Hz), 2nd-order active Butterworth low-pass filter followed by a passive low pass filter (fc = 47.5 Hz, gain = 1.61) and a passive high pass filter fc = 0.16 Hz, gain = 0.83). The filtered signals are digitized using a low-power microcontroller (MSP430F5528) with a 12-bit ADC at 512 sps, and transmitted to the CCN every 1 s at a bus rate of 100 kbps. The CCN can further transmit this data wirelessly using Bluetooth to the paired computer at a baud rate of 115.2 kbps. We have compared temporal and frequency-domain EEG signals of our system with a research-grade EEG. Results show that the proposed reconfigurable EEG captures comparable signals, and is thus promising for practical routine neurological monitoring in non-clinical settings where a flexible number of EEG channels are needed.

Sociomarkers and biomarkers: predictive modeling in identifying pediatric asthma patients at risk of hospital revisits

The importance of social components of health has been emphasized both in epidemiology and public health. This paper highlights the significant impact of social components on health outcomes in a novel way. Introducing the concept of sociomarkers, which are measurable indicators of social conditions in which a patient is embedded, we employed a machine learning approach that uses both biomarkers and sociomarkers to identify asthma patients at risk of a hospital revisit after an initial visit with an accuracy of 66%. The analysis has been performed over an integrated dataset consisting of individual-level patient information such as gender, race, insurance type, and age, along with ZIP code-level sociomarkers such as poverty level, blight prevalence, and housing quality. Using this uniquely integrated database, we then compare the traditional biomarker-based risk model and the sociomarker-based risk model. A biomarker-based predictive model yields an accuracy of 65% and the sociomarker-based model predicts with an accuracy of 61%. Without knowing specific symptom-related features, the sociomarker-based model can correctly predict two out of three patients at risk. We systematically show that sociomarkers play an important role in predicting health outcomes at the individual level in pediatric asthma cases. Additionally, by merging multiple data sources with detailed neighborhood-level data, we directly measure the importance of residential conditions for predicting individual health outcomes.

Effects of varying sampling frequency on the analysis of continuous ECG data streams

A myriad of data is produced in intensive care units (ICU) even for short periods of time. This data is frequently used for monitoring patient’s immediate health status, not for real-time analysis because of technical challenges in real-time processing of such massive data. Data storage is also another challenge in making ICU data useful for retrospective studies. Therefore, it is important to know the minimal sampling frequency requirement to develop real-time analysis on ICU data and to develop a data storage plan. In this study, we have applied the Probabilistic Symbolic Pattern Recognition (PSPR) method in Paroxysmal Atrial Fibrillation (PAF) screening problem by analyzing electrocardiogram signals at different sampling frequencies varying from 128 Hz to 8 Hz. Our results show that using PSPR method, we can obtain a classification accuracy of 82.67% in identifying PAF subjects even when the test data is sampled at 8 Hz frequency (73.33% for 128 Hz). This classification accuracy drastically improved to 92% when other descriptive features were used along with PSPR features. The PSPR’s PAF screening ability at low sampling frequency indicates its potential for real-time analysis and wearable embedded computing applications.