Tinnitus Recognition by EEG signals Based on Wavelet Transform and Deep Neural Networks

Abstract

Tinnitus seriously affects the physical and mental health of patients. Some progress has been made in the study of the electrophysiological mechanism of tinnitus. The purpose of this paper is to study the identification of tinnitus by means of EEG signal analysis. Firstly, the wavelet transform was used to extract the four frequency components of δ(0.5-3.5Hz), θ(4-7.5Hz), α(8-12Hz) and β(13-30Hz) in EEG signals. Then, the power spectrum entropy of each frequency band was calculated as the eigenvalue, and the deep neural networks (DNN) model were established to train the eigenvalues. The input layer of DNN has been a 4-dimensional eigenvector. The middle layer with two hidden layers, contained 8 neurons of each layer, in which ReLU function was adopted as activation function. In the output layer, Sigmoid function was used to classify EEG signals. Resting state EEG signals were extracted from the left middle temporal lobe of 26 subjects, and classified by three neural network models of DNN, CNN and RNN, of which the DNN with the highest classification accuracy, reaching 92%. In conclusion, there has been a certain correlation between resting state EEG signals and tinnitus, and DNN model shows a certain auxiliary diagnostic value in tinnitus recognition.