Objective Recognition of Tinnitus Location Using Electroencephalography Connectivity Features

Abstract

Purpose: Tinnitus is a common but obscure auditory disease to be studied. This study will determine whether the connectivity features in electroencephalography (EEG) signals can be used as the biomarkers for an efficient and fast diagnosis method for chronic tinnitus. Methods: In this study, the resting-state EEG signals of tinnitus patients with different tinnitus locations were recorded. Four connectivity features [including the Phase-locking value (PLV), Phase lag index (PLI), Pearson correlation coefficient (PCC), and Transfer entropy (TE)] and two time-frequency domain features in the EEG signals were extracted, and four machine learning algorithms, included two support vector machine models (SVM), a multi-layer perception network (MLP) and a convolutional neural network (CNN), were used based on the selected features to classify different possible tinnitus sources. Results: Classification accuracy was highest when the SVM algorithm or the MLP algorithm was applied to the PCC feature sets, achieving final average classification accuracies of 99.42 or 99.1%, respectively. And based on the PLV feature, the classification result was also particularly good. And MLP ran the fastest, with an average computing time of only 4.2 s, which was more suitable than other methods when a real-time diagnosis was required. Conclusion: Connectivity features of the resting-state EEG signals could characterize the differentiation of tinnitus location. The connectivity features (PCC and PLV) were more suitable as the biomarkers for the objective diagnosing of tinnitus. And the results were helpful for clinicians in the initial diagnosis of tinnitus.

Keywords: connectivity features; deep learning algorithms; objective recognition; resting-state EEG; tinnitus location.

FIGURE 1

Cortical areas are associated with tinnitus (Vanneste et al., 2018). dACC dorsal anterior cingulate cortex, sgACC subgenual anterior cingulate cortex, PCGC posterior cingulate cortex, AUD auditory cortex, PHC parahippocampus, INS insula.

FIGURE 2

Channel selection display. The highlighted channels cover the auditory cortex on the left and right sides of the brain, and these channel data were used for subsequent analysis to improve the efficiency of clinical diagnosis.

FIGURE 3

Comparison of the classification accuracy in different combinations of classifiers and features. Four machine learning algorithms were used to calculate the six feature data sets one by one to distinguish the healthy group and the three types of tinnitus patients. For the same feature set, the classification accuracy of the four models were significantly different (*p < 0.05, ^**p < 0.01, ^***p < 0.005). And the standard deviation showed that the SVM and MLP models were more stable than the CNN model.

FIGURE 4

Differences between LFT and HFT in left-sided tinnitus patients in the combination of different methods. There were some combinations that showed a significant difference between LFT and HFT (*p < 0.05, **p < 0.01, ***p < 0.005). Among the groups that showed significant differences (p < 0.05), only when the SVM-10CV combined with PLI, the recognition accuracy of HFT was lower than LFT. And the other groups were all HFT with better recognition results.

FIGURE 5

Statistical analysis of Phase lock value and Pearson correlation coefficient matrix. Both the green block and the blue block indicated that there were significant differences (p < 0.05) in the feature data between the healthy control group and the tinnitus group in this area, and the green block was the significant difference area that both existed in the three tinnitus groups.

FIGURE 6

Phase lock value and Pearson correlation coefficient matrix. PLV and PCC are both undirected connection characteristics, there is only one value between the two channels. The upper left area represents the PLV matrices and the lower right area represents the PCC matrices in each picture. Visual inspection shows clear differences in the PLV matrices and PCC matrices. The value range of PLV is from 0 to 1, indicating perfectly independent or perfectly synchronization of the two signals, respectively. PCC values of –1 and 1 correspond to perfectly negative and positive linear relationships, respectively, and a PCC value of 0 indicates that the two signals are not correlated. (A) Healthy control group. (B) Bilateral tinnitus group. (C) Left-sided tinnitus group. (D) Right-sided tinnitus group.

FIGURE 7

The data point clusters of multidimensional scaling of connectivity feature value were obtained by multidimensional cluster statistics. The highlighted area was a cluster of data points. The channels in the same cluster were highlighted with the same color in the brain topographic map, and which is in the upper right area of each picture. (1) Clusters of data points of PLV; (2) Clusters of data points of PCC.