Study on the Detection of Vertigo Induced by GVS Based on EEG Signal Feature Binary Classification

Objective: Galvanic Vestibular Stimulation (GVS) is a common method to induce reversible vertigo for the purpose of scientific or clinic study. Through the study of GVS induced vertigo, we can explore the working mechanism of vestibular nervous system. For the detection of GVS induced vertigo, we generally adopt the way of questionnaire survey of subjective feelings, which lacks objectivity. Therefore, this paper will classify GVS induced vertigo by EEG classification method to help determine GVS induced vertigo status. Methods: subjects were required to complete the Dizziness Handicap Inventory (DHI). The results were used as the data label for supervised Machine learning classification of EEG. We collected EEG signals before and after GVS stimulation. Then different sample features in EEG signals were extracted by short-time Fourier Transform, Sample Entropy (SampEn), Wavelet Transform and Ensemble Empirical Mode Decomposition (EEMD). Various machine learning classification models such as linear, nonlinear and neural network were used to classify and judge the state of vertigo. Results: The results showed that EEG classification on machine learning can realize the vertigo state detection. By comparing the classification results of various classification models under various data features, it is found that the AdaBoost nonlinear algorithm classification based on SampEn of EEG achieves the highest accuracy in the classification of whether there are vertigo, with an accuracy of 90.7% and an ROC curve area of 0.970. Conclusion: This paper obtained an ideal feature and classifier based on the feature classification of EEG signals in the detection of vertigo induced by GVS.