Evaluating Brain Activity in Response to Short Meditation Stimuli Using R-Score, Electroencephalography, and Neural Networks

Abstract

Background

Stress is a major problem in today’s culture, especially for expectant mothers. Newer therapies that show promising avenues for the reduction of stress include sound therapy, yoga, mantra chanting, and meditation (M). Electroencephalography (EEG) electroencephalography plays an important role in understanding the relaxation effects caused by these practices.

Purpose

This study looks at the immediate neuro-physiological effects of brief auditory stimuli on pregnant participants, focusing on EEG responses during and after meditation (AM). Unlike the literature that has focused on AM, the significance of EEG readings during M is emphasised.

Methods

The new classification model, namely REA (R-score, EEG and Artificial Neural Network [ANN]) was employed for the analysis against the conventional approach of ANN.

Results

The statistical feature analysis showed in M EEG more significant than AM EEG. The results indicated that short audio interventions can be used to induce relaxation, and the identified EEG channels of CFz (frontal midline), F8 (right frontal lobe), and CP2 (parietal lobe) resulted in 100% accuracy in recognising meditative states during M. This accuracy underlines the importance of these regions for the differentiation of the relaxation states.

Conclusion

The present study offers novel insights into trimester-specific responses to brief M audio stimuli and highly underscores the crucial role of EEG monitoring in M. These results point out that short auditory interventions might serve as a useful tool in the realm of stress reduction in prenatal care and offer a future direction of research into the neuro-physiological effects of M.

Introduction

In today’s lifestyle, depression and anxiety are inevitable. People are finding it harder and harder to find time for recreational activities due to being overly consumed with work-related stress and worry. For them, trying to unwind while listening to music is the best and most practical tool. Among the newer methods of relaxation is sound therapy. ¹ Engaging in yoga and meditation (M) techniques can also assist individuals in achieving mental peace. Since not everyone is an expert in these techniques, listening to these meditational sounds would be a simpler method. ² Aspects of music are used in sound healing therapy to enhance mental and physical health and well-being. Instrument playing, singing along, dancing to the beat of the music and M are a few of the activities that can be incorporated into music therapy.

There are several health advantages of M, such as reduction of stress, reduced anxiety and despair, enhanced recall, lower blood pressure, alleviation of pain, reduced cholesterol, lower chance of stroke and heart disease. ³ The scientific community is putting a lot of effort into estimating these practices’ effects, especially on the brain. Research on the brain is frequently backed by electroencephalography (EEG) data. Brain relaxation is accessible with meditative EEG. EEG records the electrical activity within the brain. With EEG, one can identify emotions, psychiatric disorders, peacefulness and any other mental state. Professionals in neurology typically record EEGs for 20 to 40 minutes, and patients ought not to be prepared for it. With the development of deep learning and machine learning techniques, this time can be reduced. Human brainwaves bear a clear mark on our emotions. Different emotional states are linked to different frequency bands in EEG data. For instance, the beta band is connected to attentiveness and focus, whereas the alpha band is linked to relaxation and serenity. There are changes in these bands during emotional stimulation to comprehend how the brain processes emotions.^4–6

Mantras and other audio stimuli, such as music, have a powerful emotional impact on humans. Positive music can produce dopamine, a neurotransmitter linked to reward and pleasure. Mantras, which are repeated chants or phrases used in M, provide the same calming and focusing effects. Over the years, auditory stimuli have been used as valuable research tools to study brain activity.^{7, 8} It has long been known that long-term auditory stimulation can cause neural reactions. Mantras are another type of audio stimulus that can be quite effective in relaxing our minds. Mantra M, which entails silently repeating a mantra while paying attention to the breath, may alter brainwave activity, according to studies. ⁹

Studies^10–12 have demonstrated a rise in alpha and theta band activity during M, suggesting a transition towards a calmer and more concentrated state of mind. Research on the relationship between EEG, emotions, auditory stimuli, and mantras is just getting started.^{13, 14} Research into the precise brain regions that are active during different emotional states and how different audio stimuli affect them can become more detailed as technology develops. This information may open the door to customised M techniques based on each person’s requirements and emotional experiences. The sections of the manuscript are broken down as follows: Section 2 explains the research conducted on contemporary works, while Sections 3 and 4 discuss the methodology chosen and the experiment’s outcomes. Section 5 contains the discussion, while Section 6 has the conclusion and future scope. The following outlines the main contributions made by this research:

1. This study introduces the use of a brief 2-minute auditory stimulus, combining the Om mantra and Tibetan bowl sounds, as an effective tool for inducing relaxation in pregnant women. This contrasts with the longer stimuli typically used in existing research.

2. The study identified specific EEG channels crucial for recognising and differentiating relaxation states both during and after meditation (AM), highlighting their importance in cognitive processes emotional processing and sensory integration.

3. A new classification model, REA, was developed and demonstrated superior performance compared to traditional ANN models, particularly in distinguishing between resting, contemplative and post-meditative stages.

4. By emphasising real-time EEG analysis during brief M sessions, the study offers fresh perspectives on the brain processes underlying mindfulness and relaxation, filling a gap in the literature that typically focuses on AM EEG recordings.

5. The study reveals the profound influence of short-term auditory stimuli on the frontal and parietal lobes, contributing to a deeper understanding of how these brain regions are engaged in the relaxation response during M.

Literature Survey

Different activities correlate with different brain states and every activity correlates with a certain EEG band. For instance, alpha is related to relaxed or awake state of mind, beta is related to thinking, delta is related to deep sleep, theta is related to deep M and gamma is related to higher mental activity. Therefore, a lot about an individual’s psychological state can be anticipated based on these bands. These bands can be used to map research on the effects of mantra as a calming activity.

Researchers examined the brainwave activity of experienced Buddhist meditators in a study. ⁷ examining the possibilities of consumer-grade EEG equipment for M research. EEG data recording during M sessions and other activities such as reading and discussion were part of the study. The researchers employed spectral analysis to look at brainwave patterns after gathering data using a consumer EEG equipment named Muse. Interestingly, the results demonstrated that, in contrast to other tasks, the delta band, which is linked to deep sleep or deep M, showed increased activity during M. Researchers used EEG in a different study ¹⁴ to look at the brain correlates of mantra M. Group C was control group sitting silently, while both Group R, M were engaged in Kirtan M. Group R consisted of 90 individuals. During a three-minute audio recording of the Hare Krishna Mahamantra chant, the Kirtan M group listened intently. This study demonstrates how promising topological data analysis (TDA) is as a novel method for examining EEG data while meditating. Although TDA and conventional band power analysis were employed for their analysis, the emphasis was on employing TDA to find new information on the topological arrangement of brain activity during various states of M.

In a different study, ¹⁵ researchers used EEG to assess the implications of yoga paired with Sudarshan Kriya (SK) M on brain activity. A practice group and a control group were created from a sample of 50 healthy male volunteers. Each participant’s 10-minute EEG sessions were recorded both prior to and following a three-month yoga and SK M intervention as component of the pre-test, post-test design utilised in the study. Remarkably, the EEG recordings were obtained following the M sessions, thereby catching possible alterations in brain activity prompted by the exercise.

Based on data taken from their EEG signals, the researchers used machine learning techniques—specifically, artificial neural network (ANN)—to differentiate meditators from non-meditators with an average accuracy of 87.2%. This study ¹⁶ investigated the precise alterations in brainwave activity brought about by NGALSO M, which combines transcendental M techniques with focused concentration. Ten experienced practitioners of M with over seven years of daily practice were gathered, along with 10 healthy controls. Using a bipolar montage, EEG recordings were made of both groups’ brain activity during a two-minute, closed-eye M session.

To evaluate alterations in brain activity, the power spectra of three distinct EEG bands—beta, alpha and theta—were examined. It’s interesting to note that, regardless of whether the measurements were taken before or AM, the skilled meditators showed significantly higher power in all three bands when compared to the controls. This implies a possible ‘trait’ impact, meaning that those who meditate on NGALSO have a unique basal neuro-physiological state with increased central nervous system activity. Additionally, the EEG waves were analysed using a machine learning approach, with a specific focus on theta bands.

The potential of EEG as a marker for expertise in NGALSO M is shown by this classification model’s impressive accuracy (varying from 80% to 90%) in differentiating experienced meditators from controls. A pilot study looked into how Om mantra M affected brainwaves, building on earlier studies on M and EEG. ¹⁷ For half an hour, 23 unskilled meditators chanted Om in the course of the research. Rather than during M, EEG recordings were made for longer than two minutes both before and after the practice. Changes in the conventional frequency bands (beta, theta, alpha and delta) were the main focus of the analysis. There was a noticeable rise in theta wave activity throughout all brain regions following M, even if no one band displayed a substantial change as a result of the practice. This result is consistent with other research that suggested theta waves are associated with a relaxed mood.

Consistent with the present study, EEG investigations incorporated in another study ¹⁸ demonstrated elevated theta band activity during M in comparison to baseline recordings. The existence of theta waves indicates a general connection between Om M and a relaxed state, even though the particular M approach (silent, chanting or listening) may vary the degree of change in different bands. Using EEG, a different study ¹⁹ looked into how Bhramari pranayama affected brain activity. Thirty experienced practitioners performed one and a half minutes of Bhramari pranayama in this study. The EEG was captured midway during the M, not after. The results showed higher average power in the right temporal area in the theta, alpha and gamma bands. These waves of the brain are linked to focused concentration and relaxation.

Using support vector machines (SVM) classification, a different study ²⁰ examined the impact of Om mantra M on the brain. For 30 minutes, Om mantras were sung by 23 novice meditators. Following M, individuals’ EEG signals were kept an eye on for two minutes as they slept with their eyes closed. The examination centred on modifications in brainwave bands. Reductions in the mean, standard deviation, variance, and interquartile range of the activity following M were most notable in the delta band (low frequency). According to this research, Om mantra M may promote a more profound level of relaxation. Particularly in the delta band, the SVM classification was able to distinguish between pre-meditation and AM EEG signals with an accuracy of 70.13%. The effects of Bhramari pranayama, a yoga breathing technique that involves humming while exhaling, on brain activity were examined in a prior study. ¹⁹

EEG was utilised in the study, which comprised 21 participants, to evaluate brain activity during M. The researchers discovered that Bhramari pranayama raised brain gamma waves, which are linked to improved cognitive function. It’s interesting to note that those who had been practicing Bhramari pranayama for a longer time showed a stronger effect of this. The latest work in this scope suggests more focus on emotional recognition and in M EEG.^20–26

The current work was inspired by a large amount of research on the effects of various mantras and meditational sounds on the human brain. The majority of the research in literature focused on long-term auditory stimulation; however, short-term auditory stimuli was not studied. Furthermore, it has been observed in literature that EEG was taken during voice M infrequently and that EEG after the M was given greater importance. But there has not been any research on EEG while M with audio stimuli.

A study ²⁷ suggests that two minutes of resting-state EEG is adequate for obtaining valuable information. Similarly, the current study recorded EEG data in all three states, each for a duration of just two minutes A two-minute audio M stimulus is used in this study to assess if it may be used to relax. The hypothesis suggests that there will be significant EEG changes during the M through short audio stimuli, demonstrating genuine relaxation. Compared to previous research, this approach focuses on EEG recordings during M. This study compares EEG collected before, during, and after short sessions of M in order to explore if these brief sessions might cause a calm state or momentary relaxation.

Methods

Using ANNs to analyse brain states before, during and AM, the study used a novel methodology that involved creating a special dataset and experimental setup. The 32-channel EEG device follows the 10–20 electrode system: Cz (1), Fz (2), Fp1 (3), F7 (4), F3 (5), FC1 (6), C3 (7), FC5 (8), FT9 (9), T7 (10), CP5 (11), CP1 (12), P3 (13), P7 (14), PO9 (15), O1 (16), Pz (17), Oz (18), O2 (19), PO10 (20), P8 (21), P4 (22), CP2 (23), CP6 (24), T8 (25), FT10 (26), FC6 (27), C4 (28), FC2 (29), F4 (30), F8 (31), Fp2 (32) as demonstrated by Figure 3. Using a 32-channel system based on the 10/20 electrode placement technique, EEG data was obtained from 43 prenatal subjects. The recordings in this dataset were made during AM, and resting states (RS). Instead of using the long-term M protocols that are commonly used, the study used a novel experimental setup that integrated short auditory stimuli as a means of inducing M, hence improving scientific rigour. Preprocessing techniques, such as strict noise reduction methods and artifact removal, were used after data gathering to guarantee data integrity. To capture the subtle EEG patterns associated with various states, the EEG signals were segmented into pertinent time epochs. Feature extraction techniques then concentrated on both spectrum analysis and time-domain properties.The EEG data were divided into pertinent time intervals, and in order to extract features that captured the subtle variations in EEG patterns linked to various states, spectral analysis and time-domain characteristics were the main focus. ANNs trained on the labelled dataset were used for classification tasks, allowing them to discriminate between the RS verses M and RS verses AM states. The effectiveness of the approach was methodically assessed by classification of M-induced relaxation states.

Figure 3. Emotiv Flex Bluetooth Headset and 10-20 Electrode System.

Participants

Forty-three expectant female participants, ranging in age from 18 to 45 years(mean age = 25 years), who had never meditated before were selected for this investigation (refer Figure 2). The subjects that were excluded from the study were critically ill patients, and patients with previous history of epilepsy. Prior to collecting data, a formal written consent form was signed by them. A short audio stimulus consisting of both the Om mantra and Tibetan bowl was employed in this investigation. The subjects were given two types of questionnaires: one to assess mental state (CESD) ²⁶ and another to assess momentary relaxation Relaxation State Questionnaire (RSQ). ²⁸ Before and after the EEG recording, the respondents were given the RSQ to see if the 2-minute audio stimulation relaxed them. After comparing the ratings before and after the experiment, it was observed that the subjects were calmer after the brief M, whereas they were depressed before the trial. This study used PEM-43 (Prenatal EEG with M) dataset, ²⁹ which is designed exclusively for pregnant women. The institutional ethics council of Gauhati Medical College and Hospital authorised the human subjects used in this investigation. The procedures followed the guidelines and standards established by the institutional ethics committee of Gauhati Medical College and Hospital. Before the EEG reading, the subjects gave their written consent. The dataset has 258 (43x2x3) samples uniformly distributed among the three classes ([RS], [M], and [AM]) and includes EEG recordings from 43 people with two trials per condition.

Figure 2. Few Participants Recording for PEM-43.

Data Collection

Forty-three expectant female participants, ranging in age from 18 to 45, who had no prior experience with M were chosen for this study. A 32-channel Emotiv Epoc Flex gel kit was used to assist in taking the EEG recording. Three states of the EEG were recorded: AM, M (while listening to audio stimuli) and RS. A 2-minute EEG was recorded by each individual in every state. Every subject had two trials. Each recording was captured at a frequency of 128 Hz. The EEG was recorded using 32 channels (with CMS/DRL references and 10–20 electrode montage) that had a dynamic range of ±4.12 mV, a resolution of 0.51 µV/bit and a range of 14 bits through a 32-channel Emotiv Epoc Flex gel kit. The EEG was internally sampled at 128 SPS (1024 Hz). The duration of the audio stimuli was set to 2 minutes, based on previous research indicating that this length is sufficient to extract useful information from EEG data. ²⁷

Using RSQ to evaluate momentary relaxation

The RSQ consists of ten items. For items 1 and 2, assign a cardiovascular score. Items 6 and 7 represent the overall level of relaxation, and items 3, 4 and 5 represent the strength rating. Items 9, 10 and 8 stand for degree of drowsiness A higher degree of relaxation is indicated by higher values for the first three scores and the scores are related. The sleepiness score and the other scores have a negative link. Higher numbers do not necessarily indicate greater relaxation; rather, they indicate increased sleepiness. The sleepy scale can therefore be used as a control variable. The individuals’ average post-relaxation score was 42.32, while their average pre-relaxation score was 21.30. A higher score indicates a state of relaxation brought on by the auditory stimulation.

Data Preprocessing

EEG signals are naturally prone to several artifacts, such as baseline drifts, muscle artifacts and eye movements (especially blinks). Slow baseline fluctuations and residual muscular movements may still be present even though the recordings in this investigation were made when subjects were closed-eyed to reduce ocular artifacts. The EEG was preprocessed using the following steps:

1. Bandpass Filtering: A Butterworth bandpass filter of fourth order Infinite Impulse Response was used, with a low-pass cutoff at 41 Hz and a high-pass cutoff at 0.5 Hz using MNE-Python toolkit. ³⁰ While the low-pass filter is especially tailored to eliminate power-line noise (usually around 50–60 Hz) and muscle activity artifacts, which tend to dominate the >40 Hz frequency range, the high-pass filter eliminates slow drifts and baseline noise. The retention of higher-frequency EEG components, such as gamma rhythms, which are important in research about M, is balanced with artifact reduction by this 41 Hz cutoff.

2. Amplitude Thresholding: Any signal amplitude above ±50 µV was removed from each EEG channel in order to further minimise contamination from eye blinks, muscular contractions, and electrode discharges. On the basis of earlier studies,^{31, 32} the amplitude cut off was determined. This threshold eliminates high-amplitude noise bursts while maintaining real brain activity because M EEG amplitudes usually lie within a smaller range. Compared to normal EEG, which can range from −100 to +100 µV, M-related EEG typically has smaller amplitudes.

3. Golay Trend Removal: Savitzky–Golay filter was used to smooth the EEG signal and eliminate baseline fluctuations or slowly changing trends following amplitude-based artifact elimination based on previous research.^{33, 34} This technique eliminates sluggish non-neural drifts while maintaining the integrity of neural oscillations by polynomial regression within a changing frame. Golay filtering, as opposed to conventional moving averages, maintains the amplitude and form of transient characteristics, which is essential for downstream analysis such as spectral feature extraction or event-related activity.

A signal’s time domain features comprise descriptive parameters such as maximum, minimum and zero-crossing rate, as well as statistical measures such as mean, variance, skewness, and kurtosis.The following statistical time-domain features were employed for feature analysis in the three states (RS, M and AM) in Table 1.

Table 1. Statistical and Frequency-Domain Features.

Feature	Description and Formula
Mean	Average of all signal values $a_i.\mu =\frac{1}{M}{\displaystyle \sum _{i=1}^M{a}_i}$
Standard deviation	Measures spread around mean; low values indicate data close to mean. $\sigma =\sqrt{\frac{1}{M}{\displaystyle \sum _{i=1}^M{\left(a_i-\mu \right)}^2}}$
Kurtosis	Describes shape, peakedness, and tail heaviness of distribution. $K=\frac{N}{M}$ where N is the fourth moment about the mean and M is variance squared.
Max	Maximum value of signal z(t): ${\max }_t(z(t))$
Min	Minimum value of signal z(t): ${\min }_t(z(t))$
Approximate entropy	Measures randomness and predictability of time series. $\begin{array}{c}AE=\frac{1}{M-(d-1)t}{\displaystyle \sum _{i=1}^{M-(d-1)t}\log }{a}_i^d(s)\\ -\frac{1}{M-(d-1)t}{\displaystyle \sum _{i=1}^{M-dt}\log }{a}_{i+1}^d(s)\end{array}$ where d = dimension, t = time delay, s = vector comparison distance.
Sample entropy	Improved Approximate Entropy without self-comparison. $SE(m,r,M)=-\ln \left(\frac{A^{m+1}}{A^m}\right)$ where A = probability of similar vector pairs within tolerance r.
Band power	Power in a frequency band by integrating the power spectral density (PSD). $P_{\text{band }}={\displaystyle {\int }_{f_{\text{low }}}^{f_{\text{high }}}\mathrm{PSD}}(f)df$

Approximate entropy and sample entropy were chosen for their capacity to quantify the complexity and irregularity of EEG-data, thereby capturing subtle changes in brain dynamics during M. Their resistance to noise and sensitivity to changing cognitive states make them excellent for studying prenatal EEG responses to brief M cues.

Classification

With their capacity to extract intricate patterns from brain waves, ANNs are being used more and more in the study of EEG-data. EEG classification and feature extraction are two areas where ANNs shine, especially deep learning designs such as convolutional or recurrent neural networks. ANNs can differentiate between various brain states, such as resting, meditating or attentive states, by interpreting the complex temporal and spatial dynamics of EEG signals. This ability is beneficial for applications such as cognitive neuroscience to brain-computer interfaces. These networks improve the accuracy of EEG-based diagnostic and cognitive evaluation tools by directly learning representations from the data, allowing them to adjust to individual variability. The three different parameters that were utilised to classify meditative states were features, channels and trimester. This classifier was chosen for the study since it has been utilised in many different signal processing investigations.^15–17 Three successive layers make up a standard ANN: the input, output and hidden layer. For each neuron j, the output is calculated as:

$$O_j=f\left({\displaystyle \sum _i{x}_{ji}}{d}_i\right)$$ (1)

where x_ji is the weight between input i and neuron j, d_i is the input signal and f is a non-linear activation function (e.g., ReLU or sigmoid).

The current study used binary cross-entropy as the loss function, which is better suitable for binary classification tasks such as differentiating between contemplative and non-meditative states, even if the above formulation frequently results in a sum-of-squares loss computation. The binary cross-entropy loss E_bce between the true label y and predicted output $\widehat{y}$ is given by:

$$E_{bce}=-\sum [y\log (\widehat{y})+(1-y)\log (1-\widehat{y})]$$ (2)

In order to increase classification performance and minimize prediction error, backpropagation is used during training to repeatedly modify weights.

The brain states during and AM (M and AM respectively) were classified in this study using EEG data analysis in contrast to the RS. ANN were used to differentiate between these states by using a dataset (PEM-43) comprising EEG recordings from 43 people over 32 channels. The ANN model employed in this study is a feedforward neural network developed for binary classification with six layers. It begins with an input layer, followed by four hidden layers containing 128, 64, 32 and 16 neurons, all of which use the ReLU activation function to introduce non-linearity. Two dropout layers are used to prevent overfitting. The final output layer contains a single neuron with sigmoid activity, making it appropriate for binary classification. To improve prediction accuracy, the model is often trained with the binary cross-entropy loss function. Figure 7 shows accuracies of channelwise classification of RS verses M and RS verses AM.

Figure 7. Accuracies of Channelwise Classification of RS vs M and RS vs AM by ANN Model.

Proposed Classification model REA

The REA model is a sophisticated tool designed to classify EEG data into different states, leveraging a combination of physiological and psychological metrics. At its core, it is an ANN architecture that integrates both EEG-data and RSQ scores. The REA model follows a sequential neural network structure with psychological scores (refer Figure 1). Thus, data flows in a linear fashion from the input layer to the output layer, passing through multiple processing layers. The input combines both EEG channel data and R scores(RSQ scores). EEG data represents brain electrical activity, while RSQ scores quantify subjective relaxation levels.The REA model is a feedforward ANN that classifies EEG data into relaxation stages by using physiological (EEG signals) and psychological (RSQ score) variables. It has a sequential architecture with an input layer that combines flattened EEG data and RSQ scores, followed by three dense layers with 128, 64 and 32 neurons, all using ReLU activation and an output layer with a single neuron using a sigmoid activation function for binary classification. To prevent overfitting, dropout layers at a 30% rate are added after each dense layer. The model is trained with the Adam optimiser with binary cross-entropy loss across fifty epochs.The REA model employs a systematic data splitting strategy to ensure reliable training and evaluation. The dataset is separated into three sets: training, validation and test. An aggregated split strategy is employed in this study, where samples from all subjects are randomly distributed across the three sets based on mediation state. Initially, 80% of the data is dedicated to training, with the remaining 20% saved for testing. The test set is then divided evenly into validation and test sets, yielding a training:validation:test ratio of 80:10:10. A binary crossentropy loss function was used in the REA model appropriate for binary classification tasks. A checkpoint method saves the best-performing model based on validation loss, assuring efficiency. The incorporation of R scores into the REA model is a novel approach that enhances the model’s ability to capture nuanced variations in EEG signals. By including subjective relaxation levels, the model gains access to psychological information that complements the physiological data from EEG. This integration allows the model to potentially identify subtle patterns in EEG that correlate with specific relaxation states.The combination of a deep neural network architecture and the integration of RSQ scores results in a powerful model capable of accurately classifying EEG data into different relaxation states. The model’s ability to learn complex patterns from EEG data, combined with the added depth provided by RSQ scores, makes it a valuable tool for understanding the relationship between brain activity and subjective relaxation experiences. In essence, the REA model offers a comprehensive approach to EEG data analysis by considering both physiological and psychological factors. The findings of this investigation demonstrate that EEG recordings obtained during M had a higher accuracy of classification than recordings obtained afterwards. In this study, EEG data analysis was employed to classify brain states during and AM (M and AM, respectively) in contrast to the RS. A new classification model, REA, was proposed to detect relaxation and was compared with ANN. The dataset (PEM-43) comprised EEG recordings from 43 individuals across 32 channels. Figure 6 presents the accuracies of channel-wise classification for RS verses M and RS verses AM. The findings reveal that EEG recordings during meditation had a higher classification accuracy than those obtained afterward. When distinguishing between the RS and M (RS vs. M), Channel 2 (Fz), located in the frontal midline, achieved a test accuracy of 100%, reflecting its crucial role in cognitive functions such as attention and working memory. Channel 31 (F8), situated in the right frontal lobe, also achieved 100% accuracy and is involved in emotional processing and executive functions. Channel 1 (Cz), at the central midline and Channel 7 (C3), in the central region, achieved accuracies of 94.44% and are critical for sensory processing and integration. Channel 3 (Fp1), found in the left frontal lobe, plays a role in higher-order cognitive processes and emotional regulation. On the other hand, for distinguishing between the RS and the AM state (RS vs. AM), Channel 27 (FC6), located in the frontal lobe, achieved 100% accuracy and is involved in changes in attention and cognitive function during meditation. These channels highlight the distinct neural alterations observed during and AM, with specific channels in the frontal lobe and central region reflecting significant changes in cognitive functions, emotional processing and sensory integration during M. The study underscores the efficacy of short-term auditory stimuli in producing quantifiable relaxation states, suggesting that even brief M sessions can significantly impact EEG patterns, especially in the parietal and frontal lobes, contributing to relaxation and mindfulness. These results highlight the unique neuronal alterations that are seen in these brain regions AM, emphasising their functions in AM moods and sensory integration processes. This suggests that temporary relaxation also affects these regions, but less so than it does during active M. These results, in contrast to the literature that focuses AM EEG recordings, support the concept that a brief 2-minute M audio stimulus is sufficient to elicit relaxation. By concentrating on the EEG evaluation of brief auditory stimuli-induced relaxation, the study closes a gap in the literature and shows that even a short M session can have a major impact on EEG patterns, especially in the parietal and frontal lobes. These results highlight the effectiveness of short-term auditory stimuli in producing quantifiable relaxation states. These findings highlight the ability of EEG-based evaluations to identify momentary relaxation, indicating that short-term audio therapies might be adequate in therapeutic settings.

Figure 1. Methodology Opted for Investigation by the Proposed Model.

Figure 6. Accuracies of Channelwise Classification of RS vs M and RS vs AM by REA Model.

In this investigation, the classification accuracy of EEG data for distinct M stages across different subjects was assessed using the RSQ in combination with the learning model. The RSQ collects people’s self-reported experiences of their level of relaxation to offer a subjective measure of relaxation. The study sought to evaluate how effectively the learning model could distinguish three different M states—RS, listening to mantra and after listening to mantra—based on these subjective relaxation ratings by merging these RSQ scores with EEG data. The EEG data was preprocessed and normalised before being used to train the machine learning model to differentiate between these states. In Figure 7 the generated accuracy scores show how well the model classified the data; they range from 0.5 to 1.0 for each subject. These accuracy ratings provided insight on the model’s capacity to distinguish between several stages of M and may indicate how well the RSQ captures pertinent characteristics linked to each condition. This method makes it possible to comprehend the relationship between objective machine learning outputs and subjective self-reports of relaxation on a deeper level. It investigates the degree to which self-reported levels of relaxation agree with the model’s predictions, offering important insights into the interaction between personal experiences and factual data analysis. The study improves our understanding of how various M states are reflected in EEG data and how precisely they may be identified based on self-reported relaxation levels by combining subjective assessments with machine learning techniques.

Results

This analysis of EEG data focuses on alterations in brainwave activity in three states: RS, M and AM in five frequency bands (Delta-δ, Theta-θ, Alpha-α, Beta-β and Gamma-γ). After dividing the EEG information into its component frequency bands—Delta-δ band, Theta-θ band, Alpha-α band, Beta-β band and Gamma-γ band—it is examined for characteristics such as Mean, Standard Deviation, Minimum, Maximum, Kurtosis, Sample Entropy, Approximate Entropy, and Band Power as shown in Figure 5. The mean values of the bands Delta-δ, Theta-θ, Alpha-α, Beta-β, and Gamma-γ are highest during the mantra listening stage (M), suggesting higher brain activity. In M, for instance, the Delta-δ band mean is 108.9369, which is more than 73.2677 in RS and 88.345 in AM. Theta-θ (97.7607), Alpha-α (73.1743), Beta-β (27.1682), and Gamma-γ bands (9.1065) likewise have their highest averages during M. This increased activity during the listening of the induced mantra indicates a higher level of emotional and cognitive engagement. Representing variability, the standard deviation exhibits a comparable trend, peaking during M across all bands. For instance, the standard deviation of the Delta-δ band in M is 2.1665, which is more than that of RS (1.4174) and AM (1.774). Enhanced brain activity and dynamic neural responses are shown by this higher variability during mantra hearing. Furthermore bolstering the heightened neural engagement theory, the minimum and maximum values peak during M. The greatest value of the theta-θ band, for instance, is 105.3549 in M, which is more than 70.88 in RS and 85.3428 in AM. This suggests that during the mantra state, there is a maximum range of brain activity. All bands show comparatively similar kurtosis values across states, indicating a constant distribution shape irrespective of the state. Kurtosis values quantify the ‘tailedness’ of the data distribution. Nonetheless, compared to RS (2.7029) and AM (2.8095), the Gamma-γ band’s kurtosis is marginally higher during M (3.2091), suggesting more severe values during mantra listening. Sample entropy is negative for the Delta-δ band during M (-0.0449) and AM (-0.0289), compared to RS (0.0142), indicating higher predictability during and after listening to the mantra. Entropy metrics are used to quantify the complexity or predictability of the EEG signal. For all bands, approximate entropy values are somewhat lower during M, suggesting simpler and more consistent patterns when listening to mantras. For all bands, band power, which is the total power inside a frequency band, reaches its maximum during M. For example, Delta-δ band power during M is 435.7477, which is greater than AM’s 353.3799 and RS’s 293.0707. This demonstrates even more how the induced mantra listening causes an increase in brain activity. As a result, the EEG data demonstrates that listening to the induced mantra greatly raises neuronal diversity and activity across all frequency bands. The mantra M appears to have a lasting effect because the post-listening state (AM) exhibits decreased but higher activity in comparison to the RS. Greater Delta-δ band values are frequently linked to states of deep relaxation and sleep, whereas greater Theta-θ band values denote calm alertness and contemplative states. Higher Gamma-γ values are connected to improved cognitive functioning and information processing, higher Beta-β band elevation is linked to active thinking and attention and increased Alpha-α activity suggests relaxation and decreased tension. When mantra listening Table 2 shows that the Delta-δ band, theta-θ band, Alpha-α band, and Beta-β band show higher mean values, indicating improved deep relaxation meditative awareness, relaxation and cognitive engagement, in that order. Deep relaxation and a contemplative state are suggested by elevated Delta-δ and theta-θ values during mantra listening, while relaxation and enhanced cognitive processing are indicated by elevated Alpha-α and Beta-β values. Indicating prolonged relaxation and cognitive involvement, post-listening scores continue to be greater than the resting condition. Chanting a mantra causes a little rise in Gamma-γ activity, which may indicate improved cognitive processes. These results reveal that the induced short mantra has a lasting effect on brain activity following a session, boosting relaxation, M and cognitive engagement.

Figure 5. Grouped Bar Chart Showing Feature Behavior Across Conditions (RS, M, AM) for Different Frequency Bands.

Table 2. Features of Each Band in Three States.

Feature	Band	RS	M	AM
Mean (in μV)	Delta-δ	73.267	108.936	88.345
	Theta-θ	65.919	97.760	79.134
	Alpha-α	49.814	73.174	58.953
	Beta-β	19.989	27.168	21.798
	Gamma-γ	8.152	9.106	8.056
Standard deviation (in μV)	Delta-δ	1.417	2.166	1.774
	Theta-θ	3.678	5.631	4.603
	Alpha-α	4.873	7.502	6.096
	Beta-β	9.030	14.036	11.139
	Gamma-γ	1.268	1.353	1.250
Min (in μV)	Delta-δ	70.869	105.274	85.342
	Theta-θ	60.199	89.002	71.975
	Alpha-α	42.688	62.194	50.040
	Beta-β	9.701	11.212	9.294
	Gamma-γ	6.517	7.310	6.43
Max (in μV)	Delta-δ	74.818	111.311	90.285
	Theta-	70.880	105.354	85.342
	Alpha-α	56.884	84.049	67.799
	Beta-β	39.243	57.078	45.725
	Gamma-γ	10.721	11.961	10.647
Kurtosis	Delta-δ	1.963	1.947	1.954
	Theta-θ	1.781	1.862	1.800
	Alpha-α	1.762	1.771	1.761
	Beta-β	2.271	2.178	2.331
	Gamma-γ	2.702	3.209	2.809
Sample entropy	Delta-δ	0.014	-0.044	-0.028
	Theta-θ	0.000	0.000	0.000
	Alpha-α	0.000	0.000	0.000
	Beta-β	0.080	0.101	0.088
	Gamma-γ	0.020	0.020	0.021
Approximate entropy	Delta-δ	0.032	-0.001	0.001
	Theta-θ	0.024	0.005	0.006
	Alpha-α	0.022	0.013	0.010
	Beta-β	0.040	0.041	0.045
	Gamma-γ	0.019	0.019	0.020
Band power (in μV)	Delta-δ	293.070	435.747	353.379
	Theta-θ	296.636	439.923	356.106
	Alpha-α	224.165	329.284	265.292
	Beta-β	349.811	475.442	381.475
	Gamma-γ	158.972	177.577	157.093

The grouped bar chart in Figure 5 depicts the comparative behavior of EEG features during RS, M, and AM circumstances for several frequency bands (Delta, Theta, Alpha, Beta and Gamma). Each feature—mean, standard deviation, minimum, maximum, kurtosis, and entropy—shows distinct tendencies among various stages. The data demonstrates that M consistently has higher feature values, notably in the Delta and Theta bands, indicating enhanced neural activity and engagement. AM values often decline when compared to the M state, but stay higher than the resting state for the majority of characteristics, indicating lingering effects of M. For example, band power and mean values in the Alpha and Beta bands show increased attention and relaxation during M, which persists afterward. These findings illustrate the dynamic changes in EEG activity caused by M, as well as its long-lasting effect in the AM period.

Figure 4 illustrates the topographic maps and shows EEG data acquired from three separate conditions. The topographic maps use a color gradient to depict the EEG voltage distribution throughout the scalp. In the plots, red and yellow regions represent higher positive amplitudes, whereas blue and cyan regions represent negative amplitudes or lower activity levels. The color transition demonstrates the spatial variance of EEG data across brain regions.The relevance of these color differences resides in their capacity to visually distinguish brain activity patterns during RS, AM and M. For example, the dominance of red in particular locations indicates increased activity or engagement, whereas blue regions indicate relaxation or inhibition of brain activity. Notably, significant alterations in frontal and parietal regions are visible during M, emphasising the engagement of critical meditative areas such as the midline, right frontal, and parietal . This color variation illustrates the differences in activation patterns caused by M and auditory stimuli, emphasising their importance in aiding relaxation and stress reduction.

Figure 4. Topography plot of Three States - RS, M, AM.

An assessment of RS verses M and RS verses AM was conducted based on trimester to determine which trimester responded best to the brief auditory stimulation. A perfect accuracy of 100% was achieved across channels 6 (FC6) 18 (FC2), 27 (FC6) and 29 (FC2) across all trimesters and an average channel accuracy of RS verses M also showed the highest accuracy of 85.34% (calculated over 50 epochs). These EEG channels, located at pivotal points such as the vertex (Cz), left frontal cortex (FC1 and FC5), occipital midline (Oz), and right frontal lobe (F8 and Fp2), are essential for cognitive processes, attention management and emotional regulation. Their consistent accuracy underscores their role in elucidating how M modulates brain activity, highlighting their significance in EEG investigations on contemplative states (refer Table 3). From this analysis, it can be concluded that EEG recorded during M holds significant value, particularly noting that participants in the third trimester responded most positively to short audio stimuli. The EEG channels showing consistent perfect accuracy across all conditions underscore their importance in understanding how M influences brain activity. This suggests that trimester-specific responses to auditory stimuli during M could offer valuable insights into cognitive and emotional processing during different stages of pregnancy.

Table 3. Classification Results of Different Conditions Based on Trimester by REA.

Trimester	Condition	Channels with 100% Accuracy
1st	RS vs. M	1, 2, 3, 6, 7, 8, 11, 15, 18, 23, 26–31
	RS vs. AM	1,3,5,6,7,8,10,13,16-32
2nd	RS vs. M
	RS vs. AM
3rd	RS vs. M	6, 10, 18, 22, 23, 26, 27, 29
	RS vs. AM

Discussion

The present investigation is in line with previous research ⁷ which reported a rise in Delta-δ band power during M in contrast to alternative activities. This discovery enhances the applicability of utilising EEG to examine M states by examining Delta-δ wave variations. By showing a drop in Delta-δ power AM as opposed to the M time itself, the current research adds to the body of knowledge. Given how brief the M period was, the results of this investigation are consistent with other studies, ¹⁴ which found that meditators had higher band power than controls.

This study builds on existing findings by showing that enhanced power is maintained throughout and AM in all bands (Delta-δ, theta-θ, Alpha-α, Beta-β and Gamma-γ). Prior research has suggested a link between theta-θ waves and a relaxed state, which is consistent with the rise in theta-θ power reported during M. ¹⁵ The use of EEG recordings during and AM in this study’s design provides a fresh interpretation of these results. Theta-θ power rose throughout both the M and AM phases as compared to the baseline, however, the classification accuracy based on EEG data during M was substantially higher than that of AM data. This implies that, as opposed to AM recordings, real-time EEG recordings made during meditation may reveal more minute variations in brain activity. This may result in increased categorization accuracy in applications such as the assessment of M states or brain-computer interfaces. Previous studies indicate a correlation between Theta-θ waves and relaxation.^{18, 19}

In contrast to conventional AM recordings, this study examined the efficacy of EEG recordings made during M for mental health assessment. Surprisingly, research indicated that EEG data obtained during M had much higher classification accuracy than baseline, despite the fact that both stages displayed higher Theta-θ power. This implies that more minute variations in brain activity are captured in real-time recordings. This research used this strategy, capturing EEG data both during and AM, after being inspired by these findings. The hypothesis that during-M recordings offer a more sensitive measure of brain activity than AM recordings is supported by the better accuracy of RS verses M in this study when compared to RS verses AM. This study supports previous research, ³⁰ which found that Theta-θ waves were higher during M than they were during RS in EEG tests. This shows that M appears to elicit a common relaxation response reflected by Theta-θ wave activity, independent of the particular technique used (silent, chanting, or listening). The majority of the literature on EEG studies of mantra M-induced calm has focused on changes in brain waves AM sessions.

But lately, various approaches have been looked upon, especially in machine learning for classification and these have proven to be far more accurate than the conventional methods. This new study outperforms previous research using SVM, with previously reported results of up to 95% accuracy. It achieves perfect 100% accuracy in trimester and channel-based categorisation, 100% accuracy in channel-based classification alone and 100% accuracy in trimester-based classification. These results represent a paradigm shift from traditional EEG approaches that rely on AM recordings, highlighting the effectiveness of brief M audio cues in promoting relaxation. This study classified the brain states during and AM (M and AM, respectively) using EEG data analysis and compared them with the RS. The novel neural networks, REA, discriminated between these states using the PEM-43 dataset, which included EEG recordings from 43 individuals across 32 channels. The results show that, in comparison to recordings made AM, EEG recordings made during M had better categorisation accuracy. Channel 2 (Fz), situated in the frontal midline, scored a 100% test accuracy when differentiating between the RS and M (RS vs. M).

This indicates its critical role in cognitive tasks including working memory and attention. Located in the right frontal lobe, Channel 31 (F8) is responsible for executive processes and emotional processing. It also attained 100% accuracy. With accuracy of 94.44%, Channel 1 (Cz), located in the centre midline, and Channel 7 (C3), located in the central area, are essential for the integration and processing of sensory data. The left frontal lobe’s Channel 3 (Fp1) is involved in higher-order cognitive functions as well as emotional control. Conversely, Channel 23 (CP2), which is involved in integrating sensory data from multiple modalities, exhibited 100% accuracy in differentiating between the RS and the AM state (RS vs. AM). This channel is located in the parietal lobe.

The parietal-temporal junction’s Channel 24 (CP6), which combines auditory and sensory inputs, has a test accuracy of 77.78%. Achieving a test accuracy of 72.22%, Channel 3 (Fp1), located in the left frontal brain, is linked to enhanced mental capacities and emotional regulation. Certain channels in the frontal lobe (Fz, F8, Fp1) and central region (Cz, C3) reveal notable changes in cognitive functioning, affective processing and sensory integration during and AM. These channels emphasise the unique neural adaptations identified during and AM. The parietal and parietal-temporal areas (CP2, CP6) contain channels that highlight the significance of sensory processing and integration in AM states.

This work highlights the potential of EEG in evaluating momentary relaxation for therapeutic purposes and advances our understanding of how brief auditory stimuli cause measurable relaxation. The study implemented a novel methodology encompassing the creation of a unique dataset and an experimental setup to investigate brain states while and AM using the proposed model REA. Using this novel approach, it was possible to analyse EEG responses to brief audio stimuli during M in a comprehensive way, providing new insights into the brain mechanisms underlying the relaxation states that are created by brief auditory treatments. The potential of customised EEG-based evaluations and individualised treatment approaches in mental health and well-being interventions is highlighted by these findings. Nevertheless, there are several limitations in this study. Only short-duration M stimuli could be used owing to the pregnant participants’ physical and physiological limitations, especially their discomfort with prolonged sitting positions. Additionally, each participant could only try two trials and additional sessions were not practical. The observed EEG patterns’ robustness and generalizability might have been limited by these variables. Further research is required to confirm and build on these findings, including larger datasets, longer M periods and repeated sessions.

Conclusion and Future Scope

This research highlights the effectiveness of short-term auditory cues in promoting relaxation, as demonstrated by notable EEG changes in numerous brain regions both during and following M, as supported by the statistical feature analysis. When compared to traditional ANN models, the innovative REA classification model performed more accurately, especially when it came to distinguishing between the resting, contemplative and AM stages. The findings imply that the EEG recorded during M is more representative of relaxation states than the recordings made later, with specific channels of the frontal and parietal lobes exhibiting the precision of the classification 100%. These findings underscore the potential therapeutic uses of auditory treatments in stress reduction, as even brief interventions can result in meaningful increases in cognitive and emotional involvement. Channel 31 (F8) in the right frontal lobe, which is important in emotional processing, Channel 23 (CP2) in the parietal lobe, which is critical for sensory integration; and Channel 2 (Fz) in the frontal midline, which is critical for cognitive functions and attentional control, all achieved 100% accuracy. Their exceptional precision reveals their crucial functions in recognising and discriminating relaxation states both during and AM, highlighting the profound influence of short-term auditory stimuli on the frontal and parietal lobes. By concentrating on real-time EEG analysis during brief M sessions, the study fills a vacuum in the literature and provides fresh insight into the brain processes underpinning mindfulness and calm. These findings highlight the potential of EEG as a tool for assessing and improving brief relaxation states and they may open the door for tailored EEG-based therapies in mental health and wellness. The study lays the groundwork for future research to explore the effects of extended M sessions using the same auditory stimuli, potentially leading to the development of more effective stress reduction techniques and personalised therapeutic interventions. The findings highlight the potential of EEG as a tool for assessing and enhancing brief relaxation states, opening the door for tailored EEG-based therapies in mental health and wellness. Future research could explore the impact of extended M sessions using the same auditory stimuli to further assess their effectiveness in inducing and sustaining relaxation. Analysing EEG data from longer M periods might offer more in-depth understanding of the neural mechanisms at play and help determine whether prolonged exposure to brief audio cues enhances or sustains the relaxation response. This could contribute to the development of more effective stress reduction techniques and personalised therapeutic interventions.

Authors’ Contribution

Daisy Das: Data curation, conceptualisation, methodology, analysis, writing-original draft, editing, Nabamita Deb: Editing, supervision, Rita Rani Talukdar: Supervision, Saswati Sanyal Choudhury: Supervision.

Data Availability

Data will be provided upon legitimate request to the corresponding author.

Statement of Ethics

Approval: The Gauhati Medical College and Hospital’s institutional ethics committee granted permission to use human subjects in research. Accordance: The methods were carried out in accordance with the relevant guidelines and regulations of Gauhati Medical College and Hospital’s institutional ethics committee.

Informed Consent

Before EEG reading, subjects signed a written consent form.

Patient Consent

All the participants signed informed consent for participation in the experiment.