Deep learning based detection and classification of vertigo induced by Galvanic Vestibular Stimulation (GVS) using EEG analyses

Abstract Vertigo is currently diagnosed either using scales as an assessment tool or via medical techniques with disadvantages that include being too subjective. Galvanic Vestibular Stimulation (GVS) is a common method used to induce reversible vertigo for scientific or clinical studies. In this study, we used GVS on participants to achieve a state of vertigo. Their pre- and post-vertigo Electroencephalography (EEG) signals were recorded using EEG. The participants completed a Dizziness Handicap Inventory (DHI) before and after GVS. The results were used as data labels for EEG deep learning classification. The aim of this study was to detect and classify the EEG signals of vertigo states to effectively detect the presence or absence of vertigo states and the severity of vertigo. A Wavelet Transform (WT) was used to remove the non-linearity and non-stationarity of the EEG signals and decompose them into different frequency bands. Automatic feature extraction of the decomposed signal is performed using Convolutional Neural Networks (CNN). This study compares the hybrid model of CNN, Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Units (GRU). The hybrid model performed better in terms of classification accuracy compared to the other models. Therefore, the hybrid model is more suitable for the diagnosis and classification of vertigo.