Machine-Deep-Ensemble Learning Model for Classifying Cybersickness Caused by Virtual Reality Immersion

Machine learning methods for the study of cybersickness: a systematic review

This systematic review offers a world-first critical analysis of machine learning methods and systems, along with future directions for the study of cybersickness induced by virtual reality (VR). VR is becoming increasingly popular and is an important part of current advances in human training, therapies, entertainment, and access to the metaverse. Usage of this technology is limited by cybersickness, a common debilitating condition experienced upon VR immersion. Cybersickness is accompanied by a mix of symptoms including nausea, dizziness, fatigue and oculomotor disturbances. Machine learning can be used to identify cybersickness and is a step towards overcoming these physiological limitations. Practical implementation of this is possible with optimised data collection from wearable devices and appropriate algorithms that incorporate advanced machine learning approaches. The present systematic review focuses on 26 selected studies. These concern machine learning of biometric and neuro-physiological signals obtained from wearable devices for the automatic identification of cybersickness. The methods, data processing and machine learning architecture, as well as suggestions for future exploration on detection and prediction of cybersickness are explored. A wide range of immersion environments, participant activity, features and machine learning architectures were identified. Although models for cybersickness detection have been developed, literature still lacks a model for the prediction of first-instance events. Future research is pointed towards goal-oriented data selection and labelling, as well as the use of brain-inspired spiking neural network models to achieve better accuracy and understanding of complex spatio-temporal brain processes related to cybersickness.

Electroencephalogram microstates and functional connectivity of cybersickness

Virtual reality (VR) is a rapidly developing technology that simulates the real world. However, for some cybersickness-susceptible people, VR still has an unanswered problem-cybersickness-which becomes the main obstacle for users and content makers. Sensory conflict theory is a widely accepted theory for cybersickness. It proposes that conflict between afferent signals and internal models can cause cybersickness. This study analyzes the brain states that determine cybersickness occurrence and related uncomfortable feelings. Furthermore, we use the electroencephalogram (EEG) microstates and functional connectivity approach based on the sensory conflict theory. The microstate approach is a time-space analysis method that allows signals to be divided into several temporarily stable states, simultaneously allowing for the exploration of short- and long-range signals. These temporal dynamics can show the disturbances in mental processes associated with neurological and psychiatric conditions of cybersickness. Furthermore, the functional connectivity approach gives us in-depth insight and relationships between the sources related to cybersickness. We recruited 40 males (24.1 ± 2.3 years), and they watched a VR video on a curved computer monitor for 10 min to experience cybersickness. We recorded the 5-min resting state EEG (baseline condition) and 10-min EEG while watching the VR video (task condition). Then, we performed a microstate analysis, focusing on two temporal parameters: mean duration and global explained variance (GEV). Finally, we obtained the functional connectivity data using eLoreta and lagged phase synchronization (LPS). We discovered five sets of microstates (A-E), including four widely reported canonical microstates (A-D), during baseline and task conditions. The average duration increased in microstates A and B, which is related to the visual and auditory networks. The GEV and duration decreased in microstate C, whereas those in microstate D increased. Microstate C is related to the default mode network (DMN) and D to the attention network. The temporal dynamics of the microstate parameters are from cybersickness disturbing the sensory, DMN, and attention networks. In the functional connectivity part, the LPS between the left and right parietal operculum (OP) significantly decreased (p < 0.05) compared with the baseline condition. Furthermore, the connectivity between the right OP and V5 significantly decreased (p < 0.05). These results also support the disturbance of the sensory network because a conflict between the visual (V5) and vestibular system (OP) causes cybersickness. Changes in the microstates and functional connectivity support the sensory conflict theory. These results may provide additional information in understanding brain dynamics during cybersickness.

Effect of Visually Induced Motion Sickness from Head-Mounted Display on Cardiac Activity

Head-mounted display (HMD) virtual reality devices can facilitate positive experiences such as co-presence and deep immersion; however, motion sickness (MS) due to these experiences hinders the development of the VR industry. This paper proposes a method for assessing MS caused by watching VR content on an HMD using cardiac features. Twenty-eight undergraduate volunteers participated in the experiment by watching VR content on a 2D screen and HMD for 12 min each, and their electrocardiogram signals were measured. Cardiac features were statistically analyzed using analysis of covariance (ANCOVA). The proposed model for classifying MS was implemented in various classifiers using significant cardiac features. The results of ANCOVA reveal a significant difference between 2D and VR viewing conditions, and the correlation coefficients between the subjective ratings and cardiac features have significant results in the range of -0.377 to -0.711 (for SDNN, pNN50, and ln HF) and 0.653 to 0.677 (for ln VLF and ln VLF/ln HF ratio). Among the MS classification models, the linear support vector machine achieves the highest average accuracy of 91.1% (10-fold cross validation) and has a significant permutation test outcome. The proposed method can contribute to quantifying MS and establishing viewer-friendly VR by determining its qualities.

Comparative Analysis of Emotion Classification Based on Facial Expression and Physiological Signals Using Deep Learning

This study aimed to classify emotion based on facial expression and physiological signals using deep learning and to compare the analyzed results. We asked 53 subjects to make facial expressions, expressing four types of emotion. Next, the emotion-inducing video was watched for 1 min, and the physiological signals were obtained. We defined four emotions as positive and negative emotions and designed three types of deep-learning models that can classify emotions. Each model used facial expressions and physiological signals as inputs, and a model in which these two types of input were applied simultaneously was also constructed. The accuracy of the model was 81.54% when physiological signals were used, 99.9% when facial expressions were used, and 86.2% when both were used. Constructing a deep-learning model with only facial expressions showed good performance. The results of this study confirm that the best approach for classifying emotion is using only facial expressions rather than data from multiple inputs. However, this is an opinion presented only in terms of accuracy without considering the computational cost, and it is suggested that physiological signals and multiple inputs be used according to the situation and research purpose.