Subject-independent Classification of Meditative State from the Resting State using EEG

TL;DR

This work tackles the challenge of subject-independent EEG classification of Rajyoga meditation versus resting state. It compares three architectures—CSP-LDA, CSP-LDA-LSTM, and SVD-NN—and demonstrates that CSP-LDA-LSTM attains up to $98.2\%$ intra-subject accuracy in the high-gamma band, while SVD-NN achieves $96.4\%$ inter-subject accuracy in the beta band, indicating strong cross-subject generalization. The study reveals frequency-band–dependent discriminability, with gamma bands bright for CSP-based methods and beta bands favorable for SVD-NN, and shows regularization or increased filter counts do not always improve performance. These findings suggest practical potential for objective, subject-independent assessment of meditation depth and state, as well as applicability to identifying other altered states of consciousness. All math-related notation is presented in $...$ format to ensure precise reporting of the methods and results.

Abstract

While it is beneficial to objectively determine whether a subject is meditating, most research in the literature reports good results only in a subject-dependent manner. This study aims to distinguish the modified state of consciousness experienced during Rajyoga meditation from the resting state of the brain in a subject-independent manner using EEG data. Three architectures have been proposed and evaluated: The CSP-LDA Architecture utilizes common spatial pattern (CSP) for feature extraction and linear discriminant analysis (LDA) for classification. The CSP-LDA-LSTM Architecture employs CSP for feature extraction, LDA for dimensionality reduction, and long short-term memory (LSTM) networks for classification, modeling the binary classification problem as a sequence learning problem. The SVD-NN Architecture uses singular value decomposition (SVD) to select the most relevant components of the EEG signals and a shallow neural network (NN) for classification. The CSP-LDA-LSTM architecture gives the best performance with 98.2% accuracy for intra-subject classification. The SVD-NN architecture provides significant performance with 96.4\% accuracy for inter-subject classification. This is comparable to the best-reported accuracies in the literature for intra-subject classification. Both architectures are capable of capturing subject-invariant EEG features for effectively classifying the meditative state from the resting state. The high intra-subject and inter-subject classification accuracies indicate these systems' robustness and their ability to generalize across different subjects.

Figures (5)

• Figure 1: Our research protocol consists of initial eyes open (IEO) and eyes closed (IEC) baselines for 3 minutes each, after which the subjects do seed stage meditation (M) for 30 minutes. The baseline recorded at the end of the experiment also has eyes closed (FEC: final eyes closed) and open (FEO) recordings for 3 minutes each.
• Figure 2: CSP-LDA-LSTM architecture which employs common spatial pattern (CSP) and long short-term memory (LSTM) for classifying meditative-state EEG epochs from the non-meditative ones. Linear discriminant analysis (LDA) is used for dimensionality reduction.
• Figure 3: SVD-NN architecture employing singular value decomposition (SVD) and shallow neural network (NN) for classifying meditative state EEG epochs from non-meditative ones.
• Figure 4: Comparison of the intra-subject classification performance of the CSP-LDA architecture for each EEG frequency band. The graph shows the results for classical CSP and TR-CSP (regularized). The y-axis gives the mean accuracy values (in %) obtained during 10-fold cross-validation. $n$ is the number of filter pairs used. The error bars show the standard deviation. The total number of meditators: 54.
• Figure 5: Comparison of the inter-subject classification performance of the CSP-LDA architecture for each EEG frequency band. The graph compares the results of classical CSP with those of TR-CSP (regularized). The Y-axis gives the mean accuracy (in %) obtained by leave-one-out cross-validation. $n$ is the number of filter pairs used. The error bars show the standard deviation. The total number of meditators: 54.