Distinguishing cognitive states using electroencephalography local activation and functional connectivity patterns

Learning and working memory are two fundamental aspects of human cognition that influence the performance of individuals during critical tasks. Several studies have demonstrated the utility of various Electroencephalography (EEG) based features in recognizing cognitive states. However, detecting cognitive states using brain activity is a challenge due to large inter-individual variability in the discriminating features. The current study investigates neural signatures of learning and working memory tasks by combining EEG local activation as well as functional connectivity patterns. In this study, EEG signals of thirty-three subjects recorded during resting, focused reading, and mental arithmetic operation were used to develop an efficient cognitive state assessment model. The feature set extracted in this study included band ratios, cognition index, entropies, and functional connectivity metrics. The optimal EEG features selected using a series of feature selection techniques were used to train six classifiers: Support Vector Machine (SVM), k-Nearest Neighbour (kNN), Linear Discriminant Analysis (LDA), Artificial Neural Networks (ANN), Ensemble, and Gaussian Naive Bayes (GNB). The fusion of various EEG features including band ratios, cognition index, entropies, coherence, and Phase-locking value yielded an average accuracy of 97.85% (Resting vs Focused Reading), 94.51% (Resting vs Mental arithmetic task), 96.98% (Focused Reading vs Mental Arithmetic task), and 88.97% (Resting vs Focused Reading vs Mental Arithmetic task) using SVM five-fold cross-validation. The proposed model can be used for automated detection of mental states during cognitive tasks such as learning, driving, air-traffic control, etc., and diagnosing cognitive-deficit diseases.