Evaluation of the Effects of Anxiety on Electro-Physiological Evoked Responses to Emotion-Related Stimuli

Abstract

Anxiety is a feeling of fear and uncertainty that can negatively affect social behavior. Understanding the impact of anxiety on social cognition is essential. This work presents an approach for the acquisition and processing of electroencephalography (EEG) signals using an emotional oddball paradigm to evaluate the effects of anxiety on electrophysiological evoked responses to emotion-related stimuli. Using two different methodologies, EEG signals were used to classify anxiety levels in 63 students. A visual emotional oddball paradigm to evoke event-related potentials (ERP) was used, and EEG signals were recorded using a commercial headset. Wavelet filtering with six levels of decomposition and reconstruction with thresholding was applied for signal processing and denoising. Subsequently, the averaged signal was obtained from 25 ERP realizations, focusing on channels O1, O2, P7, P8, F3, and F4. The first methodology used deep learning with recurrence plots of the processed signal as input for two convolutional neural networks (CNN). The second methodology used linear and nonlinear features and machine learning to classify two and three levels of anxiety. The best performance was obtained with the second method using Ramdon Forest, with an accuracy of 83.81% using a subset obtained from the wrapper method with forward selection for two anxiety levels.