A tensorial approach to access cognitive workload related to mental arithmetic from EEG functional connectivity estimates

Subject-Specific Cognitive Workload Classification Using EEG-Based Functional Connectivity and Deep Learning

Cognitive workload is a crucial factor in tasks involving dynamic decision-making and other real-time and high-risk situations. Neuroimaging techniques have long been used for estimating cognitive workload. Given the portability, cost-effectiveness and high time-resolution of EEG as compared to fMRI and other neuroimaging modalities, an efficient method of estimating an individual’s workload using EEG is of paramount importance. Multiple cognitive, psychiatric and behavioral phenotypes have already been known to be linked with “functional connectivity”, i.e., correlations between different brain regions. In this work, we explored the possibility of using different model-free functional connectivity metrics along with deep learning in order to efficiently classify the cognitive workload of the participants. To this end, 64-channel EEG data of 19 participants were collected while they were doing the traditional n-back task. These data (after pre-processing) were used to extract the functional connectivity features, namely Phase Transfer Entropy (PTE), Mutual Information (MI) and Phase Locking Value (PLV). These three were chosen to do a comprehensive comparison of directed and non-directed model-free functional connectivity metrics (allows faster computations). Using these features, three deep learning classifiers, namely CNN, LSTM and Conv-LSTM were used for classifying the cognitive workload as low (1-back), medium (2-back) or high (3-back). With the high inter-subject variability in EEG and cognitive workload and recent research highlighting that EEG-based functional connectivity metrics are subject-specific, subject-specific classifiers were used. Results show the state-of-the-art multi-class classification accuracy with the combination of MI with CNN at 80.87%, followed by the combination of PLV with CNN (at 75.88%) and MI with LSTM (at 71.87%). The highest subject specific performance was achieved by the combinations of PLV with Conv-LSTM, and PLV with CNN with an accuracy of 97.92%, followed by the combination of MI with CNN (at 95.83%) and MI with Conv-LSTM (at 93.75%). The results highlight the efficacy of the combination of EEG-based model-free functional connectivity metrics and deep learning in order to classify cognitive workload. The work can further be extended to explore the possibility of classifying cognitive workload in real-time, dynamic and complex real-world scenarios.

Modeling the Switching Behavior of Functional Connectivity Microstates (FCμstates) as a Novel Biomarker for Mild Cognitive Impairment

The need for designing and validating novel biomarkers for the detection of mild cognitive impairment (MCI) is evident. MCI patients have a high risk of developing Alzheimer's disease (AD), and for that reason the introduction of novel and reliable biomarkers is of significant clinical importance. Motivated by recent findings on the rich information of dynamic functional connectivity graphs (DFCGs) about brain (dys) function, we introduced a novel approach of identifying MCI based on magnetoencephalographic (MEG) resting state recordings. The activity of different brain rhythms {δ, 𝜃, α1, α2, β1, β2, γ1, γ2} was first beamformed with linear constrained minimum norm variance in the MEG data to determine 90 anatomical regions of interest (ROIs). A DFCG was then estimated using the imaginary part of phase lag value (iPLV) for both intra-frequency coupling (8) and cross-frequency coupling pairs (28). We analyzed DFCG profiles of neuromagnetic resting state recordings of 18 MCI patients and 22 healthy controls. We followed our model of identifying the dominant intrinsic coupling mode (DICM) across MEG sources and temporal segments, which further leads to the construction of an integrated DFCG (iDFCG). We then filtered statistically and topologically every snapshot of the iDFCG with data-driven approaches. An estimation of the normalized Laplacian transformation for every temporal segment of the iDFCG and the related eigenvalues created a 2D map based on the network metric time series of the eigenvalues (NMTSeigs). The NMTSeigs preserves the non-stationarity of the fluctuated synchronizability of iDCFG for each subject. Employing the initial set of 20 healthy elders and 20 MCI patients, as training set, we built an overcomplete dictionary set of network microstates (n μstates). Afterward, we tested the whole procedure in an extra blind set of 20 subjects for external validation. We succeeded in gaining a high classification accuracy on the blind dataset (85%), which further supports the proposed Markovian modeling of the evolution of brain states. The adaptation of appropriate neuroinformatic tools that combine advanced signal processing and network neuroscience tools could properly manipulate the non-stationarity of time-resolved FC patterns revealing a robust biomarker for MCI.

How to Build a Functional Connectomic Biomarker for Mild Cognitive Impairment From Source Reconstructed MEG Resting-State Activity: The Combination of ROI Representation and Connectivity Estimator Matters

Our work aimed to demonstrate the combination of machine learning and graph theory for the designing of a connectomic biomarker for mild cognitive impairment (MCI) subjects using eyes-closed neuromagnetic recordings. The whole analysis based on source-reconstructed neuromagnetic activity. As ROI representation, we employed the principal component analysis (PCA) and centroid approaches. As representative bi-variate connectivity estimators for the estimation of intra and cross-frequency interactions, we adopted the phase locking value (PLV), the imaginary part (iPLV) and the correlation of the envelope (CorrEnv). Both intra and cross-frequency interactions (CFC) have been estimated with the three connectivity estimators within the seven frequency bands (intra-frequency) and in pairs (CFC), correspondingly. We demonstrated how different versions of functional connectivity graphs single-layer (SL-FCG) and multi-layer (ML-FCG) can give us a different view of the functional interactions across the brain areas. Finally, we applied machine learning techniques with main scope to build a reliable connectomic biomarker by analyzing both SL-FCG and ML-FCG in two different options: as a whole unit using a tensorial extraction algorithm and as single pair-wise coupling estimations. We concluded that edge-weighed feature selection strategy outperformed the tensorial treatment of SL-FCG and ML-FCG. The highest classification performance was obtained with the centroid ROI representation and edge-weighted analysis of the SL-FCG reaching the 98% for the CorrEnv in α1:α2 and 94% for the iPLV in α2. Classification performance based on the multi-layer participation coefficient, a multiplexity index reached 52% for iPLV and 52% for CorrEnv. Selected functional connections that build the multivariate connectomic biomarker in the edge-weighted scenario are located in default-mode, fronto-parietal, and cingulo-opercular network. Our analysis supports the notion of analyzing FCG simultaneously in intra and cross-frequency whole brain interactions with various connectivity estimators in beamformed recordings.

Causal Interactions between Frontal(θ) - Parieto-Occipital(α2) Predict Performance on a Mental Arithmetic Task

Many neuroimaging studies have demonstrated the different functional contributions of spatially distinct brain areas to working memory (WM) subsystems in cognitive tasks that demand both local information processing and interregional coordination. In WM cognitive task paradigms employing electroencephalography (EEG), brain rhythms such as θ and α have been linked to specific functional roles over given brain areas, but their functional coupling has not been extensively studied. Here we analyzed an arithmetic task with five cognitive workload levels (CWLs) and demonstrated functional/effective coupling between the two WM subsystems: the central executive located over frontal (F) brain areas that oscillates on the dominant θ rhythm (Frontal^θ/F^θ) and the storage buffer located over parieto-occipital (PO) brain areas that operates on the α₂ dominant brain rhythm (Parieto-Occipital^α2/PO^α2). We focused on important differences between and within WM subsystems in relation to behavioral performance. A repertoire of brain connectivity estimators was employed to elucidate the distinct roles of amplitude, phase within and between frequencies, and the hierarchical role of functionally specialized brain areas related to the task. Specifically, for each CWL, we conducted a) a conventional signal power analysis within both frequency bands at F^θ and PO^α2, b) the intra- and inter-frequency phase interactions between F^θ and PO^α2, and c) their causal phase and amplitude relationship. We found no significant statistical difference of signal power or phase interactions between correct and wrong answers. Interestingly, the study of causal interactions between F^θ and PO^α2 revealed frontal brain region(s) as the leader, while the strength differentiated between correct and wrong responses in every CWL with absolute accuracy. Additionally, zero time-lag between bilateral F^θ and right PO^a2 could serve as an indicator of mental calculation failure. Overall, our study highlights the significant role of coordinated activity between F^θ and PO^α2 via their causal interactions and the timing for arithmetic performance.

Functional connectivity changes detected with magnetoencephalography after mild traumatic brain injury

Mild traumatic brain injury (mTBI) may affect normal cognition and behavior by disrupting the functional connectivity networks that mediate efficient communication among brain regions. In this study, we analyzed brain connectivity profiles from resting state Magnetoencephalographic (MEG) recordings obtained from 31 mTBI patients and 55 normal controls. We used phase-locking value estimates to compute functional connectivity graphs to quantify frequency-specific couplings between sensors at various frequency bands. Overall, normal controls showed a dense network of strong local connections and a limited number of long-range connections that accounted for approximately 20% of all connections, whereas mTBI patients showed networks characterized by weak local connections and strong long-range connections that accounted for more than 60% of all connections. Comparison of the two distinct general patterns at different frequencies using a tensor representation for the connectivity graphs and tensor subspace analysis for optimal feature extraction showed that mTBI patients could be separated from normal controls with 100% classification accuracy in the alpha band. These encouraging findings support the hypothesis that MEG-based functional connectivity patterns may be used as biomarkers that can provide more accurate diagnoses, help guide treatment, and monitor effectiveness of intervention in mTBI.

•

We analyzed resting state connectivity profiles in 31 mTBI patients and 55 controls.

•

We quantified frequency-specific connectivity couplings using phase-locking values.

•

Normal control networks showed dense local and sparse long-range connections.

•

TBI patient networks showed sparse local and dense long-range connections.

•

Tensor subspace analysis could classify subjects with 100% accuracy in the α band

Comparative analysis of cognitive tasks for modeling mental workload with electroencephalogram

Previous electroencephalogram (EEG) studies have shown that cognitive workload can be estimated by using several types of cognitive tasks. In this study, we attempted to characterize cognitive tasks that have been used to manipulate workload for generating classification models. We carried out a comparative analysis between two representative types of working memory tasks: the n-back task and the mental arithmetic task. Based on experiments with 7 healthy subjects using Emotiv EPOC, we compared the consistency, robustness, and efficiency of each task in determining cognitive workload in a short training session. The mental arithmetic task seems consistent and robust in manipulating clearly separable high and low levels of cognitive workload with less training. In addition, the mental arithmetic task shows consistency despite repeated usage over time and without notable task adaptation in users. The current study successfully quantifies the quality and efficiency of cognitive workload modeling depending on the type and configuration of training tasks.

Detection of variations in cognitive workload using multi-modality physiological sensors and a large margin unbiased regression machine

Physiological sensor based workload estimation technology provides a real-time means for assessing cognitive workload and has a broad range of applications in cognitive ergonomics, mental health monitoring, etc. In this paper we report a study on detecting changes in workload using multi-modality physiological sensors and a novel feature extraction and classification algorithm. We conducted a cognitive workload experiment involving multiple subjects and collected an extensive data set of EEG, ECG and GSR signals. We show that the GSR signal is consistent with the variations of cognitive workload in 75% of the samples. To explore cardiac patterns in ECG that are potentially correlated with the cognitive workload process, we computed various heart-rate-variability features. To extract neuronal activity patterns in EEG related to cognitive workload, we introduced a filter bank common spatial pattern filtering technique. As there can be large variations in e.g. individual responses to the cognitive workload, we propose a large margin unbiased recursive feature extraction and regression method. Our leave-one-subject-out cross validation test shows that, using the proposed method, EEG can provide significantly better prediction of the cognitive workload variation than ECG, with 87.5% vs 62.5% in accuracy rate.