Tinnitus EEG Classification Based on Multi-frequency Bands

Tinnitus is an auditory phantom percept of chronic high-pitched sound, ringing, or noise. Since the underlying physiological mechanisms of tinnitus are still under study, there is no universally effective treatment to cure tinnitus so far. There is even no method for objectively classifying tinnitus patients from normal people. In this paper, we utilize a Multi-view Intact Space Learning (MISL) method for the analysis and classification of electroencephalogram (EEG) signals using power value of frequency bands. At first, the power values of seven frequency bands are calculated by using Fast Fourier Transform (FFT) so as to obtain seven single views of features. Next, Multi-view Intact Space Learning is applied to integrate the seven single views together to get better classification results. Compared with the single view classification, the Multi-view Intact Space Learning method has achieved significant accuracy improvements by 6.32-23.25%. That is, the best accuracy, precision, recall and F1 of classification performance reach 0.828, 0.811, 0.857 and 0.833 respectively. The proposed method can be applied for auxiliary therapy of tinnitus as well as be extended to assist with the treatment of other diseases.