EEG-based Detection of Directed Effective Connectivity and Microstate Analysis in Stress Disorder

Abstract

Stress is a common phenomenon that affects individuals from all walks of life and it is associated with physical and physiological illness. Brain plays a key role in determining stress. This study represents the comprehensive framework for analyzing EEG data to investigate the connectivity and network dynamics under stress and normal. Our primary interest is how classifying stress in network so this is assessed through effective connectivity in which we use partial directed coherence to measure how one region influences other. Connectivity networks are computed using adjacency matrix over time and subjected to graph theoretical analysis to focus on degree and betweenness Degree highlighted highly connected region and betweenness revealed brain central region that facilitate global communication across different brain region. Further analysis in microstate that demonstrate the distinct patterns in brain, from this analysis microstate 4 show high duration and transition in normal conditions, while microstate 5 show high duration and transition in stress condition. Microstate features are used for classification into stress and normal by using three machine learning classifiers: random forest (RF), K-nearest neighbors (KNN) and support vector machine (SVM). In the classification analysis, Random Forest achieved 94% accuracy followed by KNN and SVM. Furthermost, this approach provides valuable insights into brain connectivity and demonstrate the utility of microstate features in stress classification.