Automated classification of EEG into meditation and non-meditation epochs using common spatial pattern, linear discriminant analysis, and LSTM

A deep neural network using long short-term memory (LSTM) architecture can classify EEG recordings into meditation and non-meditation segments. The approach was tested on data from fourteen long-term Rajayoga meditators. Using common spatial pattern for feature extraction and linear discriminant analysis for dimensionality reduction, the network achieved inter-subject classification accuracies of 79.1% with the alpha band, 86.5% with beta, 91.0% with lower-gamma, and 94.1% with higher-gamma bands. This is the first work to apply deep learning to distinguish brain electrical activity during meditation from resting activity.

Study at a glance

Abstract

This study proposes an approach to classify the EEG into meditation and non-meditation segments using a long short-term memory (LSTM) based deep neural network (DNN) framework. Inter-subject classification performance is assessed on EEG recorded from fourteen long-term Rajayoga meditators. Common spatial pattern is used for feature extraction, and linear discriminant analysis is used for dimensionality reduction. The sequence of features thus obtained is fed to a LSTM based DNN, which employs a fully connected layer for classification. We have achieved inter-subject classification accuracies of 79.1 %, 86.5%, 91.0%, and 94.1% with the respective use of the alpha, beta, lower-gamma, and higher-gamma bands for classification. To the best of our knowledge, this is the first work to employ deep learning to distinguish between the brain's electrical activity during meditation and at rest.