Classification of Four Mental States Using EEG Connectivity and Convolutional Neural Networks

Classifying mental states is important for identifying the type and severity of stress, in order to guide appropriate treatment. In this study, we utilized electroencephalography (EEG) signals and convolutional neural networks (CNNs) to classify four different mental states: rest, controlled alertness, stress, and stress mitigation. The control alert and stress states were induced using the Stroop color-word test (SCWT), both with and without time constraints. Stress mitigation was achieved through the use of 16-Hz binaural beat (BB) stimulation. We quantified the four mental states using target detection accuracy, subjective scores, EEG functional connectivity network images estimated by phase locking value (PLV), and CNNs. Our results showed that solving SCWT under time constraints reduced the target detection accuracy by 70% ( ${p} \lt 0.00001$ ), while BBs improved the accuracy of detection by 28% ( ${p} =0.00470$ ). The functional connectivity networks showed significant differences ( ${p} \lt 0.05$ ) between frontal/occipital and parietal regions across the four mental states. The CNNs classified the four mental states with 82.96% accuracy, 83.4% sensitivity, 94.96% specificity, 84.67% precision, and 84.03% F-score, with optimal performance at the beta band. Other EEG bands produced classification accuracies below 68%. Overall, the results indicate that 16-Hz BBS can effectively mitigate stress levels, and CNNs with beta EEG-PLV images show promise for classifying four mental states. The results of this study provide a foundation for implementing stress management interventions in order to maximize eustress and minimize distress in the workplace.