Multimodal Approach to Assess a Virtual Reality-Based Surgical Training Platform

Abstract

Virtual reality (VR) can bring numerous benefits to the learning process. Combining a VR environment with physiological sensors can be beneficial in skill assessment. We aim to investigate trainees’ physiological (ECG) and behavioral differences during the virtual reality-based surgical training environment. Our finding showed a significant association between the VR-Score and all participants’ total NASA-TLX workload score. The extent of the NASA-TLX workload score was negatively correlated with VR-Score (R2 = 0.15, P < 0.03). In time-domain ECG analysis, we found that RMSSD (R2 = 0.16, P < 0.05) and pNN50 (R2 = 0.15, P < 0.05) scores correlated with significantly higher VR-score of all participants. In this study, we used SVM (linear kernel) and Logistic Regression classification techniques to classify the participants as gamers and non-gamers using data from VR headsets. Both SVM and Logistic Regression accurately classified the participants as gamers and non-gamers with 83% accuracy. For both SVM and Linear Regression, precision was noted as 88%, recall as 83%, and f1-score as 83%. There is increasing interest in characterizing trainees' physiological and behavioral activity profiles in a VR environment, aiming to develop better training and assessment methodologies.