Classification of Mindfulness Experiences from EEG-Gamma Effective Connectivity
Poster No:
1652
Submission Type:
Abstract Submission
Authors:
Ai-Ling Hsu1, Chun-Yu Wu1, Hydra Ng2, Chi-Yun Liu3, Chih-Mao Huang4, Changwei Wu3, Chun-Hsiang Chuang2, Yi-Ping Chao1
Institutions:
1Chang Gung University, Taoyuan, Taiwan, 2National Tsing Hua University, Hsinchu, Taiwan, 3Taipei Medical University, New Taipei, Taiwan, 4National Yang Ming Chiao Tung University, Hsinchu City, Taiwan
First Author:
Co-Author(s):
Introduction:
Mindfulness refers to a transient shift of the mental process with interoceptive awareness. Previous literature has highlighted that practicing mindfulness reduces mental stress and alters the brain function or structure, along with the practice time [1,2]. However, there is lack of effective biomarkers to differentiate the mindfulness state from a mind-wandering state [3]. Based on the reported changes of electroencephalography (EEG) markers under mindfulness, a wearable neurofeedback system may enable the prediction of one's mindfulness experiences. Therefore, this study aims at probing EEG-based brain functionality to classify whether the participants had prior history of mindfulness-based stress reduction (MBSR) training.
Methods:
The dataset comprises behavioral and simultaneous EEG-fMRI recordings of 33 healthy participants aged between 20 and 80 years, same as the protocol by Ng et al [4]. All participants completed two EEG experiments, spaced eight weeks apart, and were blindly assigned to either the MBSR intervention group (n=18, mean age=47.50) or the waiting-list control group (n=15, mean age=45.87), resulting in 46 pre-intervention and 18 post-intervention EEG data. Each EEG experiment consisted of three recording sessions: resting state, focus breathing, and body scan, where each session lasted 5 minutes. Recorded 32-channel EEG data were preprocessed with standard protocol (artifact removal, bandpass filter, ICA and re-reference) of EEGLab toolbox on MATLAB 2018a. After preprocessing, gamma-band (30–40 Hz) effective connectivity between channels were calculated using the direct Directed Transfer Function (dDTF) [5]. Due to computational load, we only selected 19 channels out of 32 to calculate effective connectivity. Subsequently, WEKA workbench (v3.8.5) was employed to perform seven machine-learning analysis, and we employed principal component analysis (PCA) implemented in WEKA to transform the initial 342-dimensional data onto a new subspace. The predictive performance of each algorithm was calculated using leaving-one-out cross validation (LOOCV). Accuracy, sensitivity, and specificity were calculated to evaluate the models' performance. The statistical significance of cross-validated accuracy was set at p < 0.05, determined through a comparison with the random chance estimated by the binomial distribution (Fig.1).
·Figure 1 Before and after the 8-week MBSR training, we probed EEG effective connectivity across the 3 sessions to predict the MBSR attendance among 7 machine-learning algorithms.
Results:
Table 1 lists the performance measures of the seven classification algorithms in resting-state, focus-breathing, and body-scan sessions. The performance varied extensively across models and task sessions. Across all task sessions, the average accuracy of algorithms in the resting-state session (73.4%) outperformed those in the other two sessions of mindfulness practices (69.0% for focus-breathing and 64.3% for body-scan). Compared to sensitivity, the specificity contributed more to the accuracy in all three sessions. In the resting-state session, the algorithms of LR, SVM, NB, DT, LMT, and RF showed significantly higher accuracy of 69.4%, 75.0%, 69.4%, 91.7%, 69.4%, and 72.2%, respectively, compared to the random chance of 66.6% estimated from the binomial probability. In the focus-breathing session, four algorithms of SVM, MLP, NB, and LMT exhibited significant accuracies of 72.2%, and the RF algorithm showed a significant accuracy of 69.4%. Compared to the resting-state and focus-breathing sessions, algorithms of SVM, MLP, and NB achieved significantly higher accuracies of 72.2%, 69.4%, and 72.2% in the body-scan session above the chance level.
·Table 1: Performance measures of dDTF-based binary classifiers in three task sessions.
Conclusions:
Automatic classification of mindfulness experiences was disclosed based on the gamma-band dDTF. We found the decision tree algorithm reached the highest prediction accuracy of 91.7% with the resting state, compared to the classification accuracies of other two mindful states. By testing different algorithms and providing the effective brain features for prediction, we initiated a milestone how to objectively detect the mindfulness experience from the brain functions.
Higher Cognitive Functions:
Higher Cognitive Functions Other 2
Modeling and Analysis Methods:
EEG/MEG Modeling and Analysis 1
Novel Imaging Acquisition Methods:
EEG
Perception, Attention and Motor Behavior:
Perception and Attention Other
Keywords:
Electroencephaolography (EEG)
Machine Learning
Meta-Cognition
NORMAL HUMAN
1|2Indicates the priority used for review
Provide references using author date format
2. S.F. Santorelli, F. Meleo-Meyer, L. Koerbel, J. Kabat-Zinn, Mindfulness-Based Stress Reduction (MBSR): Authorized Curriculum Guide, Center for Mindfulness in Medicine, Health Care, and Society, 2017.
3. J. Gao, J. Fan, B.W.Y. Wu, Z. Zhang, C. Chang, Y.-S. Hung, P.C.W. Fung, H.H. Sik, Entrainment of chaotic activities in brain and heart during MBSR mindfulness training., Neuroscience Letters. 616 (2016) 218–223.
4. H.-Y.H. Ng, C.W. Wu, F.-Y. Huang, Y.-T. Cheng, S.-F. Guu, C.-M. Huang, C.-F. Hsu, Y.-P. Chao, T.-P. Jung, C.-H. Chuang, Mindfulness Training Associated With Resting-State Electroencephalograms Dynamics in Novice Practitioners via Mindful Breathing and Body-Scan, Front. Psychol. 12 (2021) 748584.
5. A. Delorme, T. Mullen, C. Kothe, Z.A. Acar, N. Bigdely-Shamlo, A. Vankov, S. Makeig, EEGLAB, SIFT, NFT, BCILAB, and ERICA: New Tools for Advanced EEG Processing, Comput. Intell. Neurosci. 2011 (2011) 130714.