EEG signal classification of tinnitus based on SVM and sample entropy

• Published in: Computer methods in biomechanics and biomedical engineering Vol. 26; no. 5; pp. 580 - 594
• Main Authors: Jianbiao, Mai, Xinzui, Wang, Zhaobo, Li, Juan, Liu, Zhongwei, Zhang, Hui, Fu
• Format: Journal Article
• Language: English
• Published: England Taylor & Francis 04.04.2023; Taylor & Francis Ltd
• Subjects: Brain; Cerebral cortex; Diagnostic systems; EEG; Electroencephalography; Electroencephalography - methods; Entropy; Humans; Pathogenesis; Sample entropy; Signal classification; Support Vector Machine; Support vector machines; SVM; Tinnitus; Tinnitus - diagnosis; Sample entropy; SVM; Tinnitus; EEG
• ISSN: 1025-5842; 1476-8259
• DOI: 10.1080/10255842.2022.2075698

The prevalence of tinnitus is high and seriously affects the daily life of patients. As the pathogenesis of tinnitus is not yet clear, there is a lack of rapid and objective diagnostic modalities. In order to provide clinicians with an objective diagnostic approach, this paper combines time-frequency domain and non-linear power analysis to investigate the differences in the specificity of the EEG signal in tinnitus patients compared to healthy subjects. In this paper, resting-state electroencephalograms (EEG) were collected from 10 cases each of tinnitus patients and healthy subjects, and the data from the two groups were compared in the δ (0.5 − 3 .5 Hz), θ (4 − 7.5 Hz), α1 (8 − 10 Hz), α2 (10 − 12 Hz), β1 (13 − 18 Hz), β2 (18.5 − 21 Hz), β3 (21.5 − 30 Hz), and γ (30.5 − 44 Hz) bands for the differences in sample entropy values. The results of the resting state experiment revealed that the δ, α2 and β1 band samples of tinnitus patients all had greater entropy values than healthy subjects, with extremely significant differences compared to healthy subjects (p < 0.01). It is mainly concentrated in the δ band in the right parietal region of the cerebral cortex, the α2 band in the central region, and the γ band in the left prefrontal region. Finally, support vector machines combined with optimal feature combinations were used to achieve objective recognition of tinnitus disorders, with an 8.58% increase in accuracy compared to other features. Through the above study, entropy reflects the degree of chaos in the brain and the chaotic characteristics of the resting state EEG signal can characterise the onset of tinnitus, the results of which can help clinicians in the early diagnosis of tinnitus.