Exploring quantitative assessment of cybersickness in virtual reality using EEG signals and a CNN-ECA-LSTM network

Abstract

Cybersickness in the virtual reality (VR) environment presents a significant challenge for user experience. Accurate evaluation methods are essential to mitigate or prevent cybersickness for the users. Compared to traditional subjective user studies such as questionnaires, electroencephalography (EEG) has the advantage of high temporal resolution, enabling more precise real-time evaluation. In this paper, we propose a novel hybrid deep learning framework that integrates a convolutional neural network (CNN), efficient channel attention (ECA) module, and long short-term memory (LSTM) to evaluate VR cybersickness using EEG signals quantitatively. A novel experimental paradigm is developed to induce cybersickness through virtual head rotation. A total of thirty-six subjects were recruited for the experiment, and both EEG signals and subjective questionnaire data were collected accordingly to build a cybersickness-EEG dataset. The proposed framework enables a meaningful evaluation of cybersickness, and the experiment demonstrates the framework’s effective ability to decode the correlation between EEG and cybersickness. The relevance between EEG signals and model predictions is also investigated to demonstrate the interpretability and effectiveness of the proposed framework. Our findings reveal a robust correlation between cybersickness and heightened activation intensity in the occipital and temporal areas.