Decoding Cognitive Load Changes Induced by Mantra Meditation From Physiological Signals Using Deep Neural Networks

Mantra meditation reduces cognitive load and improves autonomic balance, as shown by physiological markers and deep learning analysis. In a study combining EEG-derived cognitive load indices (cognitive load quotient, neural efficiency index) and heart rate variability markers (coherence index, stress index), participants who meditated showed significant reductions in cognitive load quotient and stress index, alongside increases in neural efficiency and coherence indices. Deep learning models classifying cognitive states (pre, task, post) revealed opposite accuracy trends: control group accuracy improved (EEGNet-Attention: 93.27% to 97.67%; EEGNet: 91.89% to 95.04%), while experimental group accuracy declined (EEGNet-Attention: 92.09% to 86.67%; EEGNet: 89.83% to 78.04%). This post-meditation accuracy drop suggests meditation made EEG patterns less distinct across cognitive phases, reflecting reduced cognitive load and greater neurophysiological efficiency.

Study at a glance

Abstract

Mantra meditation is widely recognized for enhancing cognition and autonomic regulation; however, its specific impact on cognitive load remains underexplored. Existing research often relies on subjective assessments or conventional electroencephalogram (EEG) spectral analysis, lacking objective, integrated evaluation frameworks. To address this gap, this study combines physiological signals including EEG-derived cognitive load indices, heart rate variability (HRV) markers, and deep learning based cognitive state classification to evaluate the effects of mantra meditation. Cognitive load was quantified using the cognitive load quotient (CLQ) and neural efficiency index (NEI), while HRV parameters such as coherence index (CI) and stress index (SI) were used to assess autonomic modulation. Deep learning models, including EEGNet and EEGNet with attention, classified cognitive states across two sessions (pre: S1 and post: S2), each comprising three phases (pre, task, and post) with continuous EEG/HRV acquisition. In the control group, accuracy improved (EEGNet-Attention: 93.27% to 97.67%; EEGNet: 91.89% to 95.04%), whereas in the experimental group it declined (EEGNet-Attention: 92.09% to 86.67%; EEGNet: 89.83% to 78.04%), showing opposite trends. The post-intervention reduction in accuracy for the experimental group suggests that meditation decreased cognitive load, leading to less distinct EEG patterns across phases. Supporting this, we observed significant reductions in CLQ and SI and increases in NEI and CI, reflecting improved cognitive efficiency and autonomic balance. Overall, the findings indicate that mantra meditation fosters neurophysiological efficiency, evidenced by both neural and autonomic markers, while reducing the separability of cognitive phases. This effect is captured through deep learning–based EEG classification.