Decoding subjective emotional arousal during a naturalistic VR experience from EEG using LSTMs

Emotional arousal (EA) denotes a heightened state of activation that has both subjective and physiological aspects. The neurophysiology of subjective EA, among other mind-brain-body phenomena, can best be tested when subjects are stimulated in a natural fashion. Immersive virtual reality (VR) enables naturalistic experimental stimulation and thus promises to increase the ecological validity of research findings i.e., how well they generalize to real-life settings. In this study, 45 participants experienced virtual rollercoaster rides while their brain activity was recorded using electroencephalography (EEG). A Long Short-Term Memory (LSTM) recurrent neural network (RNN) was then trained on the alpha-frequency (8-12 Hz) component of the EEG signal (input) and the retrospectively acquired continuous reports of subjective EA (target). With the LSTM-based model, subjective EA could be predicted significantly above chance level. This demonstrates a novel EEG-based decoding approach for subjective states of experience in naturalistic research designs using VR