Using convolutional neural networks to decode EEG-based functional brain network with different severity of acrophobia

EEG microstate in people with different degrees of fear of heights during virtual high-altitude exposure

Previous neuroimaging studies based on electroencephalography (EEG) microstate analysis have identified abnormal neural electric activity in patients with psychiatric diseases. However, the microstate information in individuals with different degrees of fear of heights (FoH) remains unknown so far. The aim of the study was therefore to explore the changes of EEG microstate characteristics in different FoH individuals when exposed to high-altitude stimulated by virtual reality (VR). First, acrophobia questionnaire (AQ) before the experiment and 32-channel EEG signals under the virtual high-altitude exposure were collected from 69 subjects. Second, each subject was divided into one of three levels of FoH including no-FoH, mild or moderate FoH (m-FoH) and severe FoH (s-FoH) groups according to their AQ scores. Third, using microstate analysis, we transformed EEG data into sequences of characteristic topographic maps and computed EEG microstate features including microstate basic parameters, microstate sequences complexity and microstate energy. Finally, the extracted features as inputs were sent to train and test an support vector machine (SVM) for classifying different FoH groups. The results demonstrated that five types of microstates (labeled as A, B, C, D and F) were identified across all subjects, of which microstates A-D resembled the four typical microstate classes and microstate F was a non-canonical microstate. Significantly decreased occurrence, coverage and duration of microstate F and transition probabilities from other microstates to microstate F in m-FoH and s-FoH groups were observed compared to no-FoH group. It was also demonstrated that both m-FoH and s-FoH groups showed a notable reduction in sample entropy and Lempel-Ziv complexity. Moreover, energies of microstate D for m-FoH group and microstate B for s-FoH group in right parietal, parietooccipital and occipital regions exhibited prominent decreases as comparison to people without FoH. But, no significant differences were found between m-FoH and s-FoH groups. Additionally, the results indicated that AQ-anxiety scores were negatively correlated with microstate basic metrics as well as microstate energy. For classification, the performance of SVM reached a relatively high accuracy of 89 % for distinguishing no-FoH from m-FoH. In summary, the findings highlight the alterations of EEG microstates in people with fear of heights induced by virtual high-altitude, reflecting potentially underlying abnormalities in the allocation of neural assemblies. Therefore, the combination of EEG microstate analysis and VR may be a potential valuable approach for the diagnosis of fear of heights.

Multi-Input CNN-LSTM deep learning model for fear level classification based on EEG and peripheral physiological signals

Objective and accurate classification of fear levels is a socially important task that contributes to developing treatments for Anxiety Disorder, Obsessive-compulsive Disorder, Post-Traumatic Stress Disorder (PTSD), and Phobia. This study examines a deep learning model to automatically estimate human fear levels with high accuracy using multichannel EEG signals and multimodal peripheral physiological signals in the DEAP dataset. The Multi-Input CNN-LSTM classification model combining Convolutional Neural Network (CNN) and Long Sort-Term Memory (LSTM) estimated four fear levels with an accuracy of 98.79% and an F1 score of 99.01% in a 10-fold cross-validation. This study contributes to the following; (1) to present the possibility of recognizing fear emotion with high accuracy using a deep learning model from physiological signals without arbitrary feature extraction or feature selection, (2) to investigate effective deep learning model structures for high-accuracy fear recognition and to propose Multi-Input CNN-LSTM, and (3) to examine the model's tolerance to individual differences in physiological signals and the possibility of improving accuracy through additional learning.

Construction and analysis of functional brain network based on emotional electroencephalogram

Networks play an important role in studying structure or functional connection of various brain areas, and explaining mechanism of emotion. However, there is a lack of comprehensive analysis among different construction methods nowadays. Therefore, this paper studies the impact of different emotions on connection of functional brain networks (FBNs) based on electroencephalogram (EEG). Firstly, we defined electrode node as brain area of vicinity of electrode to construct 32-node small-scale FBN. Pearson correlation coefficient (PCC) was used to construct correlation-based FBNs. Phase locking value (PLV) and phase synchronization index (PSI) were utilized to construct synchrony-based FBNs. Next, global properties and effects of emotion of different networks were compared. The difference of synchrony-based FBN concentrates in alpha band, and the number of differences is less than that of correlation-based FBN. Node properties of different small-scale FBNs have significant differences, offering a new basis for feature extraction of recognition regions in emotional FBNs. Later, we made partition of electrode nodes and 10 new brain areas were defined as regional nodes to construct 10-node large-scale FBN. Results show the impact of emotion on network clusters on the right forehead, and high valence enhances information processing efficiency of FBN by promoting connections in brain areas.

MuscleNET: mapping electromyography to kinematic and dynamic biomechanical variables by machine learning

Objective.This paper proposes machine learning models for mapping surface electromyography (sEMG) signals to regression of joint angle, joint velocity, joint acceleration, joint torque, and activation torque.Approach.The regression models, collectively known as MuscleNET, take one of four forms: ANN (forward artificial neural network), RNN (recurrent neural network), CNN (convolutional neural network), and RCNN (recurrent convolutional neural network). Inspired by conventional biomechanical muscle models, delayed kinematic signals were used along with sEMG signals as the machine learning model's input; specifically, the CNN and RCNN were modeled with novel configurations for these input conditions. The models' inputs contain either raw or filtered sEMG signals, which allowed evaluation of the filtering capabilities of the models. The models were trained using human experimental data and evaluated with different individual data.Main results.Results were compared in terms of regression error (using the root-mean-square) and model computation delay. The results indicate that the RNN (with filtered sEMG signals) and RCNN (with raw sEMG signals) models, both with delayed kinematic data, can extract underlying motor control information (such as joint activation torque or joint angle) from sEMG signals in pick-and-place tasks. The CNNs and RCNNs were able to filter raw sEMG signals.Significance.All forms of MuscleNET were found to map sEMG signals within 2 ms, fast enough for real-time applications such as the control of exoskeletons or active prostheses. The RNN model with filtered sEMG and delayed kinematic signals is particularly appropriate for applications in musculoskeletal simulation and biomechatronic device control.