Automated Source Estimation of Scalp EEG Epileptic Activity Using eLORETA Kurtosis Analysis

Automatic Detection and Classification of Epileptic Seizures from EEG Data: Finding Optimal Acquisition Settings and Testing Interpretable Machine Learning Approach

Deep learning (DL) is emerging as a successful technique for automatic detection and differentiation of spontaneous seizures that may otherwise be missed or misclassified. Herein, we propose a system architecture based on top-performing DL models for binary and multigroup classifications with the non-overlapping window technique, which we tested on the TUSZ dataset. The system accurately detects seizure episodes (87.7% Sn, 91.16% Sp) and carefully distinguishes eight seizure types (95-100% Acc). An increase in EEG sampling rate from 50 to 250 Hz boosted model performance: the precision of seizure detection rose by 5%, and seizure differentiation by 7%. A low sampling rate is a reasonable solution for training reliable models with EEG data. Decreasing the number of EEG electrodes from 21 to 8 did not affect seizure detection but worsened seizure differentiation significantly: 98.24 ± 0.17 vs. 85.14 ± 3.14% recall. In detecting epileptic episodes, all electrodes provided equally informative input, but in seizure differentiation, their informative value varied. We improved model explainability with interpretable ML. Activation maximization highlighted the presence of EEG patterns specific to eight seizure types. Cortical projection of epileptic sources depicted differences between generalized and focal seizures. Interpretable ML techniques confirmed that our system recognizes biologically meaningful features as indicators of epileptic activity in EEG.

Spatio-temporal dynamics of resting-state brain networks are associated with migraine disability

OBJECTIVE

The changes in resting-state functional networks and their correlations with clinical traits remain to be clarified in migraine. Here we aim to investigate the brain spatio-temporal dynamics of resting-state networks and their possible correlations with the clinical traits in migraine.

METHODS

Twenty Four migraine patients without aura and 26 healthy controls (HC) were enrolled. Each included subject underwent a resting-state EEG and echo planar imaging examination. The disability of migraine patients was evaluated by Migraine Disability Assessment (MIDAS). After data acquisition, EEG microstates (Ms) combining functional connectivity (FC) analysis based on Schafer 400-seven network atlas were performed. Then, the correlation between obtained parameters and clinical traits was investigated.

RESULTS

Compared with HC group, the brain temporal dynamics depicted by microstates showed significantly increased activity in functional networks involving MsB and decreased activity in functional networks involving MsD; The spatial dynamics were featured by decreased intra-network FC within the executive control network( ECN) and inter-network FC between dorsal attention network (DAN) and ECN (P < 0.05); Moreover, correlation analysis showed that the MIDAS score was positively correlated with the coverage and duration of MsC, and negatively correlated with the occurrence of MsA; The FC within default mode network (DMN), and the inter-FC of ECN- visual network (VN), ECN- limbic network, VN-limbic network was negatively correlated with MIDAS. However, the FC of DMN-ECN was positively correlated with MIDAS; Furthermore, significant interactions between the temporal and spatial dynamics were also obtained.

CONCLUSIONS

Our study confirmed the notion that altered spatio-temporal dynamics exist in migraine patients during resting-state. And the temporal dynamics, the spatial changes and the clinical traits such as migraine disability interact with each other. The spatio-temporal dynamics obtained from EEG microstate and fMRI FC analyses may be potential biomarkers for migraine and with a huge potential to change future clinical practice in migraine.

A review of MEG dynamic brain network research

The dynamic description of neural networks has attracted the attention of researchers for dynamic networks may carry more information compared with resting-state networks. As a non-invasive electrophysiological data with high temporal and spatial resolution, magnetoencephalogram (MEG) can provide rich information for the analysis of dynamic functional brain networks. In this review, the development of MEG brain network was summarized. Several analysis methods such as sliding window, Hidden Markov model, and time-frequency based methods used in MEG dynamic brain network studies were discussed. Finally, the current research about multi-modal brain network analysis and their applications with MEG neurophysiology, which are prospected to be one of the research directions in the future, were concluded.

Combination of Group Singular Value Decomposition and eLORETA Identifies Human EEG Networks and Responses to Transcranial Photobiomodulation

Transcranial Photobiomodulation (tPBM) has demonstrated its ability to alter electrophysiological activity in the human brain. However, it is unclear how tPBM modulates brain electroencephalogram (EEG) networks and is related to human cognition. In this study, we recorded 64-channel EEG from 44 healthy humans before, during, and after 8-min, right-forehead, 1,064-nm tPBM or sham stimulation with an irradiance of 257 mW/cm². In data processing, a novel methodology by combining group singular value decomposition (gSVD) with the exact low-resolution brain electromagnetic tomography (eLORETA) was implemented and performed on the 64-channel noise-free EEG time series. The gSVD+eLORETA algorithm produced 11 gSVD-derived principal components (PCs) projected in the 2D sensor and 3D source domain/space. These 11 PCs took more than 70% weight of the entire EEG signals and were justified as 11 EEG brain networks. Finally, baseline-normalized power changes of each EEG brain network in each EEG frequency band (delta, theta, alpha, beta and gamma) were quantified during the first 4-min, second 4-min, and post tPBM/sham periods, followed by comparisons of frequency-specific power changes between tPBM and sham conditions. Our results showed that tPBM-induced increases in alpha powers occurred at default mode network, executive control network, frontal parietal network and lateral visual network. Moreover, the ability to decompose EEG signals into individual, independent brain networks facilitated to better visualize significant decreases in gamma power by tPBM. Many similarities were found between the cortical locations of SVD-revealed EEG networks and fMRI-identified resting-state networks. This consistency may shed light on mechanistic associations between tPBM-modulated brain networks and improved cognition outcomes.