The effect of brief and focused mindfulness meditation on young adults using EEG approach with machine learning

doi:10.21203/rs.3.rs-4962094/v1

Our study seeks to investigate the neural mechanisms underlying brief mindfulness meditation through the utilization of electroencephalogram (EEG) recordings and machine learning techniques. Twenty-four participants underwent mindfulness training, with pre- and post-practice physiological assessments and EEG recordings. Deep learning was used to analyze EEG data to predict mindfulness states. Models like Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) classified meditation vs. rest, with LSTM achieving 72.8% accuracy, MLP 62.4%, and CNN 64.1%. In addition, EEG analysis revealed that reduced frontal delta activity and increased alpha, beta, and gamma activity during a brief and focused mindfulness meditation. The power spectral density (PSD) of focused mindfulness meditation state in the left (F7) and right frontal (F4, TP10) was positively correlated with difference in heart rate before and after meditation. Our results in EEG and physiological measurements suggested that brief and focused mindfulness meditation could reduce anxiety and stress by inducing a hypometabolic state of the body and a tranquil but alert state of the mind. The LSTM model is a reliable data-driven approach to predict the state of focused mindfulness meditation.

focused mindfulness meditation

deep learning

long short-term memory (LSTM)

electroencephalogram (EEG)

Mindfulness requires individuals to intensely focus on their sensations and feelings in the present moment, without interpretation or judgment [1]. Meditation is a cognitive enhancement method that facilitates brain remodeling [2]. Mindfulness meditation usually combines meditation with the practice of mindfulness, which entails full attention to the present moment and the acceptance of thoughts, feelings, and sensations without judgment [3].

Mindfulness meditation can be categorized into two types: open mindfulness meditation and focused mindfulness meditation [4, 5]. Open mindfulness meditation involves a broader observation of things around oneself, the body, and the mind (e.g., thoughts, feelings, memories) with an open and vairagya (non-attachment) approach [4]. On the other hand, focused mindfulness meditation involves concentrating on a specific object (e.g., breathing, repeated spells, or imaginary images) [5].

Both open and focused mindfulness meditation offer valuable benefits. However, compared with open mindfulness meditation, focused mindfulness meditation provides a clear and tangible point of focus (e.g., breathing). Thus, focused mindfulness can make it easier for the beginners to understand and practice as they have a specific anchor to return to when the mind wanders during mindfulness practices [4, 6]. Studies have shown that focused mindfulness meditation can enhance focused attention more effectively than open mindfulness meditation [6]. Moreover, staying focused on the present moment allows individuals to observe their thoughts and emotions without immediate judgment, leading to reduced reactivity and increased resilience in the face of stress [7]. Recent studies have demonstrated that focused mindfulness meditation improves emotional regulation, including the reduction of negative emotions and enhancements in physical and mental health [8].

As mindfulness meditation receives increasing attention, brief focused mindfulness meditation is gradually obtaining public acceptance. Brief focused mindfulness meditation is a shorter meditation practice, typically lasting from a few minutes to ten minutes [9]. Such practice allows individuals to quickly reap the benefits of meditation amidst their busy daily lives, promoting relaxation of the mind and body, reducing stress, and enhancing concentration. Research has shown that even a brief mindfulness meditation practice can positively impact mood regulation and cognitive functioning [10]. Therefore, our study would primarily investigate the effects of brief and focused mindfulness meditation in young adults.

As the scientific community increasingly strives to quantify the effects of mindfulness meditation, various methods have been employed to establish neural association in mindfulness meditation. One such technique is Electroencephalography (EEG), which records physiological electrical brain activity and provides measurements of large-scale neural network synchronization. EEG is used to analyze brain activity in specific meditative states [11]. EEG data is usually analyzed in different frequency bands, namely theta, alpha, beta, and gamma waves, with frequency bands of 4–7 Hz, 8–13 Hz, 14–30 Hz, and 31–80 Hz respectively.

Traditional approach of EEG analysis is usually hard to empirically elucidate the nonlinear, multidimensional, and complex system-level endogenous electrophysiological activity that likely characterizes the intricate brain states during mindfulness meditation [12]. Deep learning (DL) is a type of machine learning that utilizes computational methods to learn intrinsic patterns or characteristics from samples. Recently, several researchers have applied deep learning to EEG data from mindfulness meditation [13–16] by Support Vector Machine (SVM) and a spectral feature-based Artificial Neural Network (ANN) to classify the mindfulness meditation expertise of focused breathing practitioners. Sharma et al. [17] used Artificial Neural Networks (ANN) to distinguish between meditators and non-meditators.

Three models of deep learning might be applied in classification of EEG data features: the Multilayer Perceptron (MLP), the Long Short-Term Memory (LSTM), and the Convolutional Neural Network (CNN). MLP is a type of feedforward neural network consisting of an input layer, hidden layers, and an output layer. It utilizes the backpropagation algorithm for network parameter training. Due to its simple structure and ease of training, MLP was widely employed in early studies on the classification of brain electroencephalogram (EEG) signals [18]. However, MLP exhibits limited capabilities in handling time-series signals and fails to effectively extract temporal dynamic features from EEG signals. LSTM is a type of recurrent neural network, incorporates memory units and gate mechanisms in its hidden layers [19], making it more suitable for processing time-series data. In EEG signal classification tasks, LSTM performs well [20] because EEG signals inherently possess temporal dependencies, and can capture long-term temporal relationship. CNN employs convolutional and pooling layers to extract local and spatial features from input signals [21]. In processing EEG signals, CNN can learn spatial characteristics between different channels [22] but exhibits weaker capabilities in modeling temporal dynamic information.

Previous EEG studies not only identified a number of mindfulness meditation-related EEG frequencies in theta and alpha wave activity [23], but also showed that mindfulness meditation is longitudinally associated with increased theta and alpha band power of the EEG [24, 25]. After the intervention of mindfulness, the alpha and theta wave power of the patient's brain area increased. The change in power fluctuation was more obvious on the left side of the brain area than on the right side, and the brain waves on both sides showed asymmetric changes. Furthermore, different types of mindfulness meditation practices are associated with unique frequency patterns [26].

Additional studies have shown that alpha and theta wave power in the brain tends to increase during mindfulness meditation compared to the resting state [27]. This simultaneous increase in alpha and theta waves may suggest a relaxed state of alertness, which is beneficial for mental health [11]. When compared to a resting state, mindfulness meditation activates activity in the frontal lobes, promoting relaxation and reducing an individual's anxiety, heart rate, and stress levels [27]. Previous research has also indicated that reduced blood flow to the superior temporal lobe, cingulate, and prefrontal lobes is associated with increased alpha wave power. This suggests that the brain consumes less energy, inducing a more relaxed state and leading to increased positive emotions and feelings of calm [26, 28].

The autonomic nervous system (ANS) is a major organizing component of emotional experience, and highly cognitive neural states affect the ANS [29]. Heart rate (HR), heart rate variability (HRV), and respiratory rate (RR) are commonly used as biomarkers of the ANS [30]. Mindfulness meditation has different physiological effects on heart rate, respiratory rate, blood pressure, skin conductance, and alpha oscillations. This suggests that mindfulness meditation induces a state of physiological stillness [31].

Previous research has explored the concussive changes of overall mindfulness meditation, whereas the potential mechanisms of focused mindfulness meditation training still need further exploration. Additionally, machine learning could be used to classify the EEG data features to illustrate more electrophysiological properties of the brain instead of traditional EEG analysis [32]. Our objective was to investigate EEG data characteristics during distinct states of focused mindfulness meditation compared to resting states utilizing deep learning techniques. We additionally conducted a comparative analysis of EEG signals and physiological parameters between the baseline resting state and the focused mindfulness meditation state. Ultimately, we examined the relationship between the power spectral density of EEG signals during focused mindfulness meditation and specific electrophysiological indices (such as heart rate and blood pressure) during the states of focused mindfulness meditation.

2.1 Methods

2.1.1 Participants

In this study, 29 college students from Shanghai University of Traditional Chinese Medicine were recruited. Five participants were excluded (two participants fell asleep, three participants had poor EEG data quality or technical error). The final sample for analysis included 24 participants, 14 females (58.33%) and 10 males (41.67%), with an average age of 20.13 ± 1.85 (N = 24). None of them had any mindfulness meditation or other meditation experience (Table 1). All participants were recruited through inclusion and exclusion criteria as the following rules. All subjects signed informed consent forms prior to the experiment and was paid for their services. The study was in accordance with the Declaration of Helsinki and approved by Yueyang Hospital of Integrated Traditional Chinese and Western Medicine affiliated to Shanghai University of Traditional Chinese Medicine (2020 − 178).

Inclusion criteria: (1) Full-time undergraduate students aged 18–22. (2) No prior participation in or self-study of any meditation practices. (3) No language or communication barriers. (4) Provided informed consent for this study and complied with ethical standards. Exclusion criteria: (1) History of major cardiovascular, hepatic, renal, pulmonary, hematologic, gastrointestinal, endocrine, immunologic, dermatologic, or neurological diseases before the study. (2) Ingestion of any alcoholic or caffeinated foods or beverages within 48 hours before the mindfulness meditation session. (3) History of severe psychiatric comorbidities other than anxiety and stress (e.g., bipolar disorders, psychotic disorders) or substance use disorder. (4) Noncompliance with the experimental arrangement. All participants were recruited via flyers that invited healthy volunteers to participate in a study of mindfulness meditation. All of the participants were right-handed and had normal or corrected-to-normal vision. Informed consent was obtained from the participants before the study, and the participants received some payment after the experiment.

2.2 Measure

2.2.1 EEG recording and processing

In this study, EEG recordings were conducted using a 32-channel non-invasive saline electrode cap manufactured by Greentek Pty Co., Ltd. (Wuhan, China). The placement of electrodes followed international standards as illustrated in Fig. 1. The Smarting Pro EEG amplifier, developed by mBrainTrain Co., Ltd. (Serbia), was utilized for recording EEG signals at a sampling frequency of 500 Hz and a data resolution of 24 bits [33]. Throughout the EEG recording session, strict measures were taken to maintain electrode impedance below 20kΩ. Subsequently, the acquired data was wirelessly transmitted in real-time via Bluetooth to the Smarting Pro desktop application (mbtStreamer). The EEG data was meticulously recorded and analyzed across distinct frequency bands, encompassing 1–4 Hz (delta), 4–8 Hz (theta), 8–13 Hz (alpha), and 13–30 Hz (beta). Two reference electrodes, specifically A1 (left ear) and A2 (right ear), were employed—one on each earlobe—to ensure accurate signal referencing. Subsequently, the examination environment was meticulously controlled to maintain optimal conditions, characterized by a tranquil setting at room temperature, appropriate lighting, and minimal visual and auditory disturbances. The acquisition of patient EEG data is conducted by skilled engineers adhering to stringent operational protocols and a standardized workflow.

The operator placed the non-gel electrode cap on the subject's head, ensuring that the Cz electrode was located at the intersection of the midline of the forehead and the line connecting the tips of the ears. The transmitter of the EEG amplifier was connected to the non-gel electrode cap, and the receiver of the EEG amplifier was inserted into the USB port of the computer. The wet Link Electrode cotton was installed into the electrode to make it contact the scalp. The Smarting Streamer acquisition software was opened to connect the EEG amplifier to the computer software, and the subject's brain waves were measured. The entire experiment process needed to be kept quiet, and the subject was asked to minimize movement during EEG measurement.

After the measurement was completed, the Link Electrode cotton was first removed, followed by the EEG cap. The amplifier was dismantled. The Link Electrode cotton in the holder was taken out one by one, washed in clean water, squeezed dry, repeated three times, and then dried and placed in a box for easy use next time. The electrode cap was washed with clean water, dried in the shade, and placed. During the cleaning process of the EEG cap, avoid touching the wires.

2.2.2 Physiology recording

2.2.2.1 Blood pressure and heart rate

YE660D arm-type electronic sphygmomanometer produced by Yuyao Company was used to measure the participants' blood pressure and heart rate (pulse rate) after resting state and mindfulness meditation. Studies have shown that pulse rate in young people can be an effective substitute for heart rate [34, 35].

2.2.2.2 Respiratory rate

The operator observed the number of thoracic movements (ups and downs) for 30 seconds, and then calculated the respiratory rate by multiplying the count by 2.

2.3 Mindfulness meditation training

The 10-minute mindfulness exercise for this study was an audio recording of a focused mindfulness meditation by a guide with mindfulness teacher qualifications. The content of the guides was derived from the course exercises of Mindfulness-based cognitive therapy for life (MBCT-L) in the Oxford Center for mindfulness. The intervention plan was as follows: First, invite the participants to prepare their bodies and environment, sensing the surroundings, relaxing their muscles, and softening their gaze. Feel the weight of the entire body, staying present in the feeling of sitting quietly in this moment. This process lasts for 2 minutes. Next, invite the participants to focus their attention on their breath, choosing a comfortable spot where breathing is easily felt. Feel the changes in airflow, speed, temperature, and humidity with each breath, noticing the temperature and speed of breath entering the nostrils, again and again. This process lasts for 7 minutes. Finally, gradually shift your attention from the breath to the body, feeling its stability. When ready, open your eyes. This process lasts for 1 minute.

To mitigate evaluator bias, interventions were recorded by an author qualified as a mindfulness instructor. Audio interventions were administered by two additional authors, while intervention assessments were conducted by two other authors. Consistent evaluations by the two assessors yielded the final assessment for the participants. In cases of disagreement, discussion with a third assessor was undertaken to reach a final evaluation.

2.4 Experimental procedure

(1) After entering the test room, participants sit still for 5 minutes to rest fully.

(2) The operator collected the participants' blood pressure, heart rate, and respiratory rate.

(3) Started the EEG measurement.

(4) The participants were asked to rest for 10 minutes.

(5) The participants were asked to practice meditation for 10 minutes.

(6) Completed the EEG measurement.

(7) The operator collected the participants' blood pressure, heart rate, and respiratory rate again.

2.5 Data analysis

2.5.1 Deep learning analysis

In this study, a cohort of 24 participants was recruited, and each participant's brain activity was recorded using 32 EEG channels, covering a span of 2050 discrete time points, during both the resting and focused mindfulness meditation states. To prepare the dataset, the time series data from the 32 EEG channels of randomly selected 22 participants were concatenated. The corresponding labels, indicating whether the brain activity corresponded to a resting or focused mindfulness meditation state, were also appended to each time point in the concatenated dataset. Subsequently, this combined dataset was partitioned into training and validation sets using a random allocation strategy with an 8:2 split ratio. Furthermore, the recorded EEG data from the remaining 2 participants were deliberately withheld and utilized as unseen data, reserved for the purpose of evaluating the model's performance as a test set.

In the context of input preparation for the deep learning model, the initial 2D signal array undergoes a transformation to 3D format. This conversion is achieved by introducing an additional dimension, which serves the purpose of delineating the distinct channels within the data, encompassing time steps. The conversion process is as follows Table 1:

Table 1

Dimension transformation process of original data
Model		Original X	Expand X
MLP	-	(m, n)	(m, n)
LSTM	np. expand_dims(X, axis = 0)	(m, n)	(1, m, n)
CNN	np. expand_dims(X, axis = 2)	(m, n)	(m, n, 1)

¹ Where n represents each channel and m represents the time step.

The study involved a comparative analysis of three deep learning models, specifically the MLP, LSTM, and CNN. The proposed integrated deep learning frameworks in this study productively combine complementary sub-model advantages. Notably, it is essential to highlight that normalization was not utilized during the preprocessing stage, as empirical experimentation revealed that the inclusion of normalized data did not yield an improvement in the model's overall accuracy.

2.5.2 EEG analysis

The EEG data from 24 participants underwent preprocessing and analysis utilizing software in conjunction with the EEGLAB v2023.0 toolbox. The procedural steps are outlined as follows:

The EEGLAB toolbox within MATLAB was utilized to derive channel locations based on the provided experimental channel location file. Subsequently, a bandpass filter (1–90 Hz) was applied, followed by referencing through an average reference scheme. Segmentation of the resting state and focused mindfulness meditation state EEG signals into 2-second segments was performed based on markers recorded during data acquisition using MATLAB software. To exclude ocular, muscle activity and other artifacts, Independent Component Analysis (ICA) was employed by using EEGLAB routines. Segments containing excessive noise were identified and removed using the extreme value method, with a threshold set at ± 80µV. Finally, a whole-brain average re-referencing procedure was executed on the EEG signal.

In this study, a total of 32 channels were effectively utilized for data collection. Each participant contributed 15 random 2-second segments (equating to 45 seconds of data) for the computation of functional connectivity. The signal was segmented into four distinct frequency bands: δ (1-4Hz), θ (4-8Hz), α (8-13Hz), and β (13-30Hz), enabling the assessment of functional connectivity across varying frequency ranges. Spectrum analysis, a fundamental technique in EEG quantification, involves decomposing the intricate EEG signal into its constituent frequencies via the application of the Fourier transform. The power spectrum, indicative of activity levels within specific frequency bands, is derived through this process [36].

The power spectral density (PSD) was calculated using the pwelch function, following the Welch mean period method [37]. Frequency bands of 0–45 Hz and 55–90 Hz were chosen to exclude EEG power frequency interference [38]. A comparison of the spectra between the resting state and meditation state was conducted, and a paired t-test was performed on the frequency points of the 32 electrode channels individually. Additionally, the correlation between PSD and changes in heart rate and respiratory rate during meditation was analyzed using the spCorrDotLine function. The test level was set at α = 0.05, and statistical significance was considered for p < 0.05.

2.5.3 Physiology analysis

In this study, we utilized SPSS 25.0 to analyze physiological differences, including blood pressure, heart rate, and respiratory rate, before and after focused mindfulness meditation. The normality of the data was ensured prior to the analysis, then we performed paired t-tests to analyze the results of the two tests before and after the focused mindfulness meditation.

2.5.4 Correlation analysis

By using Pearson correlation, we investigated whether there was a correlation between the channel PSD of focused mindfulness meditation and these electrophysiological measures (heart rate, blood pressure) for the practitioners during focused mindfulness meditative states.

3.1 Demographic data

As shown in the participant flow diagram for the study (Fig. 2). The demographic characteristics of the participants are shown in Table 2.

Table 2

**The demographic characteristics of the participants**
Characteristics	Information	Rate
Gender	female	14 (58.33%)
Gender	male	10 (41.67%)
Age	-	20.13 ± 1.85
Education	college	21 (87.5%)
Education	other	3 (12.5%)
Area	city	18 (75%)
Area	countryside	6 (25%)
Meditation experience	no	24

3.2 The accuracy of prediction by deep learning for focused mindfulness meditation

The performance of deep learning models was rigorously assessed through comprehensive testing. Findings revealed that the LSTM model achieved the highest accuracy on the validation set, reaching 74.1%. On the independent test set, the LSTM model demonstrated a notable accuracy of 72.8%, whereas the MLP and CNN models achieved accuracies of 62.4% and 64.1%, respectively (refer to Fig. 3 and Table 3).

Table 3

Total parameters and accuracy of each model in test set
Model	Params	Accuracy (Validation/Test)	TPR (Validation/Test)	TNR (Validation/Test)	AUROC (Validation/Test)
MLP	4,225	0.67 / 0.62	0.64 / 0.59	0.72 / 0.78	0.81 / 0.84
LSTM	19,009	0.74 / 0.73	0.70 / 0.76	0.78 / 0.76	0.83 / 0.91
CNN	33,217	0.66 / 0.64	0.64 / 0.73	0.71 / 0.74	0.78 / 0.88
Ensemble	56,451	0.79 / 0.77	0.78 / 0.76	0.79 / 0.76	0.84 / 0.92

3.3. Data of EEG analysis

After conducting paired t-tests, the analysis revealed significant differences between the resting state and focused mindfulness meditation state in the following frequency bands: PSDs in the δ-band of FP2 (right temporal pole of the prefrontal region) (1.70898 Hz ~ 2.44141 Hz), and PSDs in the α-band (12.207 Hz ~ 12.4512 Hz), β-band (19.5312 Hz ~ 26.3672 Hz), and γ-band (31.4941 Hz ~ 38.8184 Hz, 42.7246 Hz ~ 43.7012 Hz) of the FZ (frontal zone frontal midline). In addition, significant differences were also observed in the 80.0781 Hz to 81.0547 Hz frequency band of the CP1 channel (Fig. 4).

In the δ-band of right temporal pole of the prefrontal region, the PSD during the resting state surpassed that of focused mindfulness meditation, indicating a consistent decline as frequency increased. Conversely, within the frontal midline and parietal regions of the frontal region, the PSD in these bands during focused mindfulness meditation exceeded that during the resting state, displaying a consistent decrease with increasing frequency. The brain topography is illustrated in Fig. 5.

3.4 Blood pressure, heart rate, and respiratory rate

The result revealed that following focused mindfulness meditation, there was a significant decrease in both heart rate (68.50 ± 9.17 vs 73.54 ± 11.10; p < 0.001) and respiratory rate (14.83 ± 4.83 vs 17.75 ± 4.47; p < 0.001) compared to pre-meditation levels of heart rate. However, there was no significant changes observed in systolic and diastolic blood pressure levels pre- and post-focused mindfulness meditation. These results are presented in Fig. 6 and Table 4.

Table 4

Dimensionality transformation process of raw data
Physiological indexes	Pre-meditation	Post-meditation	t	p
Systolic blood pressure	111.79 ± 14.75	107.67 ± 15.91	1.76	0.09
Diastolic blood pressure	70.04 ± 8.35	67.650 ± 8.01	1.50	0.15
Heart rate	73.26 ± 11.10	68.50 ± 9.17	4.22	< 0.001
Respiratory rate	17.75 ± 4.47	14.83 ± 4.83	5.05	< 0.001

3.5 Exploratory correlational analysis

The correlation coefficient between the change in respiratory rate and the change in PSD of F4 (right frontal) was positive: r = 0.48 (p < 0.05) during the mindfulness meditative state (Fig. 7). Moreover, the correlation coefficients between the change in heart rate and the difference in PSD of F4 (right frontal), F7 (left frontal), and TP10 measures were positive: r = 0.55 (p < 0.05), 0.60 (p < 0.01), and 0.43 (p < 0.05) respectively during the mindfulness meditative state (Fig. 7). However, the correlation coefficients between the change in blood pressure and the difference in PSD of any channels were not significant (p < 0.01) during the focused mindfulness meditative state.

Brief and focused mindfulness meditation is commonly utilized as an adjunctive therapy for psychological or psychosomatic conditions. In this study, deep learning emerges as a dependable data-centric methodology for forecasting the intricate systemic neuro-electrophysiology associated with brief and focused mindfulness meditation. The MLP, LSTM, and CNN models were utilized to fuse their individual outputs into prediction results for the test dataset. The results revealed the ensemble model prediction accuracy (79.0%) for brief and focused mindfulness meditation, implying that the discernibility of mindfulness meditative states from alternative mental states grounded on electrophysiological activity. Furthermore, two key discoveries emerged in the present study. Primarily, in contrast to the resting condition, brief and focused mindfulness meditation state induced a progressive decrease in the frontal delta activity, alongside an elevation in frontal alpha, beta, and gamma activity. Secondly, the PSD of the focused mindfulness meditation state within the left (F7) and right frontal (F4, TP10) regions exhibited a positive with the heart rate differential (mindfulness meditative state vs. resting state). These observations hold potential significance in elucidating the underlying neural mechanisms associated with short-time, focused mindfulness mediation practice.

4.1 The modeling accuracy of brief and focused mindfulness meditation using deep learning

The ultimate ensemble model demonstrated an accuracy of 79.0% on the validation set and 77.2% on the test set in the context of focused mindfulness meditation, representing notable enhancements of 4.6% and 4.4%, correspondingly, compared to the LSTM model's performance. Prior studies have shown a classification accuracy (78%-90%) in identifying meditative states [13–15, 39]. The investigation elucidates that each of the three deep learning models, MLP, LSTM, and CNN, exhibits unique strengths and limitations. In real-world applications, a holistic approach incorporating CNN for spatial feature extraction, coupled with either LSTM or MLP for temporal information processing, shows promise in attaining enhanced representation learning for EEG signals. The results of our investigation indicate that hybrid models, created through the fusion of diverse architectures, can outperform stand alone models by capitalizing on their synergistic capabilities. The integration of MLP, LSTM, and CNN models into a unified entity exhibited significant performance advantages over individual architecture. Subsequent research efforts should prioritize the investigation of optimal fusion frameworks to effectively harness the synergistic potential of these models.

4.2 The sensitive EEG bands for focused mindfulness meditation

The study reveals that in contrast to the resting state, the state of focused mindfulness meditation exhibits elevated Power Spectral Density (PSD) in the alpha, beta, and gamma bands across the frontal region. These outcomes align with earlier research findings [8, 15, 38, 40, 41]. Numerous studies have documented the correlation between mindfulness meditation and PSD within the alpha band, reflecting a state characterized by relaxation intertwined with vigilance [4, 11, 13, 41, 42]. Research conducted by Cahn et al. demonstrated that proficient meditators display increased alpha power levels during meditation in contrast to both control groups and individuals without meditation experience [43]. Likewise, Lomas et al. suggested that mindfulness meditation is associated with heightened alpha power relative to baseline resting conditions [14], implying a state of tranquil alertness conducive to profound relaxation and well-being. These results suggest that the focused mindfulness meditation state is marked by a heightened state of alert relaxation compared to the resting state, facilitating individuals in attaining a deeper sense of calmness and contentment [13]. Additionally, focused mindfulness meditation could potentially signify an optimal state for relaxation [42].

Furthermore, beta waves are the predominant common high-frequency waves typically detected during states of wakefulness [43]. They play a pivotal role in enabling precise movements during conscious activities including cognitive reasoning, calculation, reading, communication, and cognitive processes [44]. The study unveiled an elevation in beta power within the prefrontal lobes immediately following the commencement of the meditation session [43]. Some researchers attribute this observation to the diminished state of high arousal and heightened levels of attention and relaxation noted in practitioners of mindfulness meditators [45].

Thirdly, two lower gamma frequency bands (31.49 Hz to 38.82 Hz; 42.72 Hz to 43.70 Hz) and three higher gamma bands (44.43 Hz to 48.34 Hz) demonstrated an increase during focused mindfulness meditation in contrast to the resting state. Prior studies have established a robust association between mindfulness meditation and Gamma waves [5, 11, 38]. Lutz et al. detected prominent gamma oscillations with high-amplitude gamma oscillations in the EEG recordings of experienced meditators during the meditative state, which were not present in the initial baseline [5]. Additionally, Wang et al. pointed out that the clustering coefficient and global efficiency of the gamma network exhibited an augmentation during the deep meditation subphase [38]. Neural synchrony, specifically within the gamma-band frequencies (35-80Hz), is commonly linked to cognitive functions such as attention, working memory, cognitive learning, and conscious perception [45]. One possible explanation for the rise in gamma power may be attributed to the elevated frequency of EEG oscillations; enhancements in prefrontal and gamma power were observed immediately after the commencement of meditation and persisted throughout the meditation session [38].

Interestingly, in comparison to the resting state, the focused mindfulness meditation state exhibited a reduction in delta power along the frontal. This result aligns with previous studies [39, 44]. Delta power is usually associated with mind wandering [37]. This finding suggests that, relative to resting state, the focused mindfulness meditation state is associated with reduced mind wandering, a characteristic that aligns with prior research indicating a connection between mindfulness mediation practice and attentional control [46].

4.3 The physiological associated with focused mindfulness meditation

The study revealed a decline in both heart rate and respiratory rate following focused mindfulness meditation in comparison to pre-focused mindfulness meditation levels. Nevertheless, no significant alterations in blood pressure were noted pre- and post-meditation. Previous research have demonstrated that practices such as breathing meditation can effectively lower an individual's heart rate [47, 48] and respiratory rate [49]. Lumma et al. documented a reduction in heart rate among seasoned meditation practitioners during breathing meditation [48]. Intensive Vipassana retreats are linked with unique respiratory alterations that are leveraged to enhance concentration [47]. However, our results diverged from with those reported by Walsh et al. Specifically, Walsh et al.'s investigation revealed no significant variation in heart rate within the MT group when contrasted with the control group. Discrepancies from Walsh et al.'s outcomes could potentially stem from experimental conditions that prompted heightened heart rates prior to engaging in focused meditation. Furthermore, Walsh et al.'s study implies that under real-world conditions (e.g., outside the laboratory), the practice of focused meditation may not lead to a substantial decrease in heart rate, and in some instances, could even result in an increase[48, 50, 51].

The positive correlation observed between the average difference in power across frontal channels (F4, F7, TP10) and the change in heart rate before and after mindfulness meditation aligns with findings from prior studies [45, 52]. Shanok et al.'s research revealed that decreases in depression levels from pre-to-post intervention were mediated by enhancements in beta and alpha power [52]. The fluctuations in heart rate and respiratory rate observed during mindfulness meditation could be attributed to the activation of the relaxation response. Focused mindfulness meditation elicits activation of the parasympathetic nervous system, which governs the body's 'rest and digest' functions. Through directing attention towards the breath or a specific focal point, the body transitions into a state of relaxation, countering the 'fight or flight' response of the sympathetic nervous system [48]. This activation of the relaxation response leads to a decrease in heart rate and breathing rate. The stability in blood pressure levels may be attributed to the nuanced impact of focused mindfulness meditation on blood pressure, which is contingent upon the duration and frequency of practice. Transient or sporadic practice may not yield in consistent or significant changes in blood pressure. Sustained and regular practice is likely requisite to discern more pronounced effects on blood pressure. Collectively, the proficiency in focused mindfulness meditation mirrors for emotion regulation, exerting influence on both physiological and neural responses [52].

4.4 Limitation and future work

There were some noticeable limitations that need to be considered. Firstly, the limited sample size, leading to substantial inter-subject variability that might not be effectively captured by DL. Subsequent studies endeavors could derive advantages from enlarged sample cohorts to allocate greater unseen proportions, thereby enhancing the approximating to real-world deployment. Secondly, the absence of a control group in this study hindered the ability to make robust comparison, minimizes bias, bolster causal inferences, and enhance generalizability of the findings[53]. Subsequent investigations necessitate the inclusion of a control group to authenticate the outcomes. Thirdly, this study exclusively scrutinized EEG distinctions between focused mindfulness meditation and the resting state using frequency domain analysis. Wang et al. identified a propensity for alterations in low-frequency oscillations during the states of light meditative, whereas high-frequency oscillations exhibit a higher likelihood of modification during deeper meditative stages [39]. However, this study did not ascertain the precise intrinsic correlation between individual frequencies and varying degrees of light, medium, and deep mindfulness meditation. Future studies endeavors could delve deeper into diverse states of focused mindfulness meditation through a time-frequency perspective. Finally, mindfulness meditation encompasses focused meditation and open meditation practices, as well as long-term and short-term modalities. Future research could delve into the EEG and physiological mechanisms underlying extended durations of focused and open meditation practices utilizing deep learning methodologies. It is pertinent to acknowledge that this study employed a single session of the mindfulness intervention, and investigating the psychological and neurological adaptations associated with sustained focused mindfulness states across multiple sessions could constitute a key area of interest for future investigations.

In summary, we investigated the neural mechanisms underpinning focused mindfulness meditation. Delta, alpha, beta, and gamma are identified as sensitive EEG bands associated with focused mindfulness meditation and the focused mindfulness meditation serves as a modality for inducing hypometabolic state in the body, fostering a tranquil yet cognitively alert mental state conducive to relaxation. Additionally, three deep learning models were utilized for EEG data classification, with the LSTM model exhibiting superior accuracy. It demonstrated notable efficacy in discerning in distinguishing the EEG data features unique to mindfulness meditation. Deep learning holds promise as a potent instrument for investigating the neural physiological substrates and underlying brain mechanism of mindfulness meditation.

Author Contributions: Conceptualization, J.L. and Y.L.; data curation, YQ.Z. and J.L.; investigation, Y.L. and S.X.; methodology, Y.C., W.H., Y.Q., and Z.Z.; software, YY.Z. and Z.L.; validation, N.A., L.H. and Z.L.; formal analysis, YY.Z. and L.S.; supervision, P.S. and J.L.; writing—original draft preparation, YY.Z.; writing—review and editing, Y.L. and YQ.Z. All authors have read and agreed to the published version of the manuscript. (Yaoyao Zhang, Yongxin Luo, Yuqing Zhaicontributed equally to this work.)

Funding: This research was funded by Shanghai Municipal Commission of Economy and Informatization (grant number2023-GZL-RGZN-01012) and Postdoctoral Fellowship Program of CPSF (GZB20230600).

Institutional Review Board Statement: The study involving human participants were reviewed and approved by ChiCTR2000040609.

Informed Consent Statement: Not applicable.

Data Availability Statement: The data and code presented in this study are available upon request from the corresponding author.

Acknowledgments: We would like to thank the direct support professionals who participated in this study. Additionally, we thank students for her help on this project.

Conflicts of Interest: The authors declare no conflict of interest.

Kabat-Zinn, J.C. Mindfulness-based interventions in context: past, present, and future. Clin Psychol: Science and Practice2003,10, 144-156.
Kozas, E.H.; Balardin, J.B.; Ricardo, S.J. Effects of a 7-Day meditation retreat on the brain function of meditators and non-meditators during an attention task. Front Hum Neurosci 2018, 12, 222.
Tang, Y.Y.; Tang, R.; Posner, M.I. Mindfulness meditation improves emotion regulation and reduces drug abuse. Drug Alcohol Depend2016,163, 13-18.
Cahn, B.R.; Polich, J.X. Meditation states and traits: EEG, ERP, and neuroimaging studies. Psychol Bull 2006, 132, 180-211.
Lutz, A.; Slagter, H.A.; Dunne, J.D.; Davidson, R.J. Attention regulation and monitoring in meditation. Trends Cogn Sci 2008, 12, 163-169.
Edenfield.; Teresa, M.; Saeed, S.Y.; Atezaz. An update on mindfulness meditation as a self-help treatment for anxiety and depression. Psychol Res Behav Manag2012.
Menezes, C. B.; de Paula Couto, M. C.; Buratto, L.G.; Erthal, F.; Pereira, M.G.; Bizarro, L. The Improvement of Emotion and Attention Regulation after a 6-Week Training of Focused Meditation: A Randomized Controlled Trial. Evid Based Complement Alternat Med2013,2, 1-11.
Deng, X.; Yang, M.; An, S. Differences in frontal EEG asymmetry during emotion regulation between high and low mindfulness adolescents. Biol Psychol2021,158, 107990.
Zhang, Y.; Bao, X.; Yan, J.; Miao, H.; Guo, C. Anxiety and depression in Chinese students during the COVID-19 pandemic: A meta-analysis. Front Public Health 2021,9, 697642.
Kaplan, S.; Berman, M.G. Directed attention as a common resource for executive functioning and self-regulation. Perspect Psychol Sci2010,5, 43-57.
Lomas, T.; Ivtzan, I.; Fu, C.H.Y. A systematic review of the neurophysiology of mindfulness on EEG oscillations. Neurosci Biobehav Rev2015,57, 401-410.
Schoenberg, P.L.; Vago, D.R. Mapping meditative states and stages with electrophysiology: concepts, classifications, and methods. Curr Opin Psychol2019,28, 211-217.
Deolindo, C.S.; Ribeiro, M.W.; Aratanha, M.A.; Afonso, R.F.; Irrmischer, M.; Kozasa, E.H. A critical analysis on characterizing the meditation experience through the electroencephalogram. Front Syst Neurosci 2020, 14, 1-29.
Hagad, J.L.; Fukui, K.; Numao, M. Deep visual models for EEG of mindfulness meditation in a workplace setting. Precision Health Med: A Digital Revolution in Healthcare2020, 129-137.
Kora, P.; Meenakshi, K.; Swaraja, K.; Rajani, A.; Raju, M S. EEG based interpretation of human brain activity during yoga and meditation using machine learning: A systematic review. Complement Ther Clin Pract 2021,43, 101329.
Lee, D.J; Kulubya, E.; Goldin, P; Goodarz, I.A.; Girgis, F. Review of the neural oscillations underlying meditation. Front Neurosci2018,12, 178.
Sharma, H.; Ra, J.R.; Juneja, M. EEG signal-based classification before and after combined Yoga and Sudarshan Kriya. Neurosci Lett 2019,707,134300.
Keehn, B.; Linke, AC.; Perez-Edgar, K. Impacts of attention-deficit/hyperactivity disorder on electrophysiological markers of attention allocation. Biological Psychiatry: Cognitive Neurosci Neuroimage2021,6, 489-498.
Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput1997,9, 1735-1780.
Schirrmeister, R.T.; Springenberg, J.T.; Fiederer, L.D.; Glasstetter, M.; Eggensperger, K.; Tangermann, M.; Ball, T. Deep learning with convolutional neural networks for EEG decoding and visualization. Hum Brain Mapp2017, 38, 5391-5420.
LeCun, Y.; Boser, B.; Denker, J.S.; Henderson, D.; Howard, R. E.; Hubbard, W.; Jackel, L.D. Backpropagation applied to handwritten zip code recognition. Neural Comput1989,1, 541-551.
Bashivan, P.; Rish, I.; Yeasi, M.; Codella, N. Learning representations from EEG with deep recurrent-convolutional neural networks. Comput Sci2015,16, 1063-6919.
Craik, K.A.; He, Y.; Contreras-Vidal, L.J. Deep learning for electroencephalogram (EEG) classification tasks: A review. J Neural Eng2019, 16.
An, A.; Hoang, H.; Trang, L.; Vo, Q.; Tran, L.; Le, T.; Le, A.; McCormick, A.; du Old, K.; Williams, N.S.; Mackellar, G.; Nguyen, E.; Luong, T.; Nguyen, V.; Nguyen, K.; Ha, H. Investigating the effect of mindfulness-based stress reduction on stress level and brain activity of college students. IBRO Neurosci Rep2022,12, 399-410.
Morais, P.; Quaresma, C.; Vigario, R.; Quintao, C. Electrophysiological effects of mindfulness meditation in a concentration test. Med Biol Eng Comput 2021, 59, 759-773.
Lagopoulos, J.; Xu, J.; Rasmussen, I.; Vik, A.; Malhi, G.S.; Eliassen, C. F.; Arntsen, I.E.; Saether, J.G.; Hollup, S.; Holen, A.; Davanger, S.; Ellingsen, Ø. Increased theta and alpha EEG activity during nondirective meditation. J Altern Complement Med 2009,5, 1187-1192.
Fell, J.; Axmacher, N.; Haupt, S. From alpha to gamma: Electrophysiological correlates of meditation-related states of consciousness. Med Hypotheses2010,75, 218-224.
Cahn, B.R.; Delorme, A.; Polich, J. Occipital gamma activation during vipassana meditation. Cogn Process2010,11, 39-56.
Anon. Maternal and neonatal complications of substance abuse in Iranian pregnant Women. Acta Medica Iranica2012,50, 411-416.
Michael, T.B.; Wilson, M.C. Does physical activity protect against drug abuse vulnerability?. Drug Alcohol Depend2015, 3-13.
Manuello, J.; Vercelli, U.; Nani, A.; Costa, T.; Cauda, F. Mindfulness meditation and consciousness: an integrative neuroscientific perspective. Conscious Cogn2016,40, 67-78.
Jo, T.; Hou, J.; Eickholt, J.; Cheng, J. Improving protein fold recognition by deep learning networks. Sci Rep 2015,5, 17573.
Jakovljević, T.; Janković, M.M.; Savić, A.M.; Soldatović, I.; Todorović, P.; Jere Jakulin, T.; Papa, G.; Ković, V. The sensor hub for detecting the developmental characteristics in reading in children on a white vs. colored background/colored overlays. Sensors (Basel, Switzerland)2021,21, 406.
Bolanos, M.; Nazeran, H.; Haltiwanger, E. Comparison of heart rate variability signal features derived from electrocardiography and photoplethysmography in healthy individuals. Annu Int Conf IEEE Eng Med Biol Soc2006,4289-4294.
Shi, P.; Hu, S.J.; Zhu, Y.S. A preliminary attempt to understand compatibility of photo plethysmographic pulse rate variability with electrocardiogramic heart rate variability. J Med Biol Eng 2008,28, 173-180.
Sun, P.; Zhang, S.; Jiang, L.; Ma, Z.; Yao, C.; Zhu, Q.; Fang, M. Yijinjing Qigong intervention shows strong evidence on clinical effectiveness an electroencephalography signal features for early poststroke depression: A randomized, controlled trial. Front Aging Neurosci2022, 14, 956316.
Dasilva, M.D.; Gonçalves, Ó.F.; Branco, D.; Postma, M. Revisiting consciousness: Distinguishing between states of conscious focused attention and mind wandering with EEG.Conscious Cogn 2022,101, 103332.
Wang, M.; Guo, M.; Wang, X.; Xue, T.; Wang, Z.; Li, H.; Guo, X. Dynamic alterations of EEG oscillations and networks from resting to deep meditation. In 2019 12th International Congress on Image and Signal Processing. Bio Medical Engineering and Informatics (CISP-BMEI)2019, 1-6.
Aviad, N. Oscillating mindfully: using machine learning to characterize systems-level electrophysiological activity during mindfulness meditation (Doctoral dissertation). University of Haifa (Israel)2021.
Qiu, J.; Xu, J.; Qian, D.; Qu, W.; Zhang, X.; Mao, W. Preliminary investigation on effect of mindfulness therapy on clinical efficacy and EEG in patients with somatoform disorders. J Psychiatr 2018,31, 340-343.
Tong, H.; Liu, L.; Liu, L.; Qi, S.; Wang, Z. On psychological and EEG of siting in forgetfulness mindfulness, and meditation anxiety therapy. Med Philos 2017,38, 84-85.
Ahani, A.; Wahbeh, H.; Nezamfar, H. Quantitative change of EEG and respiration signals during mindfulness meditation. J Neuroeng Rehabil 2014,11, 87.
Cahn, B.R.; Polich, J. Meditation states and traits: EEG, ERP, and neuroimaging studies. Psychol Bull, 2013,32, 180-211.
Symons, A.E.; El-Deredy, W.; Schwartze, M.; Kotz, S.A. The functional role of neural oscillations in non-verbal emotional communication. Front Hum Neurosci2016,10, 239.
Kim, D.K.; Rhee, J.H.; Kang, S.W. Reorganization of the brain and heart rhythm during autogenic meditation. Front Integr Neurosci 2014,13, 109.
Gan, Q.; Ding, N.; Bi, G.; Liu, R.; Zhao, X.; Zhong, J.; Chen, Z. Enhanced resting-state functional connectivity with decreased amplitude of low-frequency fluctuations of the salience network in mindfulness novices. Front Hum Neurosci2022,16, 838123.
Krygier, J.R.; Heathers, J.A.; Shahrestani, S.; Abbott, M.; Gross, J.J.; Kemp, A.H. Mindfulness meditation, well-being, and heart rate variability: a preliminary investigation into the impact of intensive Vipassana meditation. Int J Psychophysiol 2013,89, 305-313.
Lumma, A.L.; Kok, B.E.; Singer, T. Is meditation always relaxing? Investigating heart rate, heart rate variability, experienced effort and likeability during training of three types of meditation. Int J Psychophysiol2015,97, 38-45.
Oliva, V.; Baumgartner, J.N.; Farris, S.R.; Riegner, G.; Khatib, L.; Jung, Y.; Zeidan, F. Neural and psychological mechanisms in the relationship between resting Breathing Rate and Pain. Mindfulness2023,1-10.
Walsh, K.M.; Saab, B.J.; Farb, N.A. Effects of a mindfulness meditation app on subjective well-being: Active randomized controlled trial and experience sampling study. JMIR Ment Health2019,6, e10844.
Johnston, J.M.; Minami, T.; Greenwald, D.; Li, C.; Reinhardt, K.; Khalsa, S.B. Yoga for military service personnel with PTSD: A single arm study. Psychol Trauma: theory, research, practice and policy2015, 7, 555–562.
Shanok, N.A.; Saldias-Manieu, C.; Mize, K.D.; Chassin, V.; Jones, N.A. Mindfulness-training in preadolescents in school: The role of emotionality, EEG in theta/beta bands, creativity and attention. Child Psychiatry Hum Dev2023,54, 1152-1166
Emanuel, E.J.; Wendler, D.; Grady, C. What makes clinical research ethical. Jama2000,283, 2701-2711.

No competing interests reported.

The effect of brief and focused mindfulness meditation on young adults using EEG approach with machine learning

Status:

Version 1

Abstract

Figures

1. Introduction

2. Materials and Methods