A deep learning approach for analyzing brainwave signals during audio meditation in expectant mothers: variational mode decomposition with a CNN-BiLSTM model.

Abstract

Meditation has increasingly been recognized as a helpful non-pharmacological intervention to treat psychological stress, anxiety, and depression during pregnancy-a period that plays a significant role in maternal and fetal health. Although very popular, there is very little scientific evidence on how brief audio meditation affects brain activity in pregnant women. This work reports an EEG-based method to objectively investigate the neural effect of short-term audio meditation in pregnant women with the aid of sophisticated signal decomposition and deep learning methods. Electroencephalogram (EEG) signals were acquired from pregnant women in various mental states: resting, meditation, and post-meditation. For improved feature representation, Variational Mode Decomposition (VMD) was used to decompose each EEG channel into five intrinsic mode functions, representing multi-resolution signal components. The IMFs were subsequently processed through a hybrid deep learning model integrating Convolutional Neural Networks (CNNs) for spatial feature extraction and Bidirectional Long Short-Term Memory (BiLSTM) networks for temporal dynamics modeling. The test accuracy of the recommended VMD-CNN-BiLSTM architecture was 94.79%, and test loss was 0.2975, proving its high reliability in classifying EEG states during meditation. The model's performance verifies its ability to extract frequency-specific and temporal features characteristic of short-term cognitive modulation. This study offers a new, objective framework to evaluate the effect of meditation during prenatal care and opens avenues toward personalized, brain-based mindfulness training in maternal care.