Deep Learning End-to-End Approach for the Prediction of Tinnitus based on EEG Data. Annu Int Conf IEEE Eng Med Biol Soc 2021;2021:816-819. [PMID: 34891415 DOI: 10.1109/embc46164.2021.9629964]

Use of Some Relevant Parameters for Primary Prediction of Brain Activity in Idiopathic Tinnitus Based on a Machine Learning Application

INTRODUCTION

Tinnitus is one of the most common complaints, distressing about 15-24% of the adult population. Because of its pathophysiology heterogeneity, no curable treatment has been attained yet. Even though a neuromodulation management technique based on the tinnitus network model is currently being developed, it has not yet worked because the most involved brain areas still remain unpredictable from the patient's individual clinical and functional profile. A remarkable correlation between tinnitus network activity and the subjective measures of tinnitus like perceived loudness and annoyance and functional handicap is well established. Therefore, this study aimed to develop software for predicting the involved brain areas in the tinnitus network based on the subjective characteristics and clinical profile of patients using a supervised machine-learning method.

METHODS

The involved brain areas of 30 tinnitus patients ranging from 6 to 80 months in duration were recognized by using QEEG and sLORETA software. There was a correlation between subjective information and those areas of activities in all rhythms by which we wrote our software.

RESULTS

For verification and validation of the software, we compared and analyzed the results with SPSS data and the receiver operating characteristic (ROC) curves.

CONCLUSIONS

The findings of this study confirmed the effectiveness of the software in predicting the brain activity in tinnitus subjects; however, some other important parameters can be added to the model to strengthen its reliability and feasibility in clinical use.

Deep learning-based electroencephalic diagnosis of tinnitus symptom

Tinnitus is a neuropathological phenomenon caused by the recognition of external sound that does not actually exist. Existing diagnostic methods for tinnitus are rather subjective and complicated medical examination procedures. The present study aimed to diagnose tinnitus using deep learning analysis of electroencephalographic (EEG) signals while patients performed auditory cognitive tasks. We found that, during an active oddball task, patients with tinnitus could be identified with an area under the curve of 0.886 through a deep learning model (EEGNet) using EEG signals. Furthermore, using broadband (0.5 to 50 Hz) EEG signals, an analysis of the EEGNet convolutional kernel feature maps revealed that alpha activity might play a crucial role in identifying patients with tinnitus. A subsequent time-frequency analysis of the EEG signals indicated that the tinnitus group had significantly reduced pre-stimulus alpha activity compared with the healthy group. These differences were observed in both the active and passive oddball tasks. Only the target stimuli during the active oddball task yielded significantly higher evoked theta activity in the healthy group compared with the tinnitus group. Our findings suggest that task-relevant EEG features can be considered as a neural signature of tinnitus symptoms and support the feasibility of EEG-based deep-learning approach for the diagnosis of tinnitus.

Side-Aware Meta-Learning for Cross-Dataset Listener Diagnosis With Subjective Tinnitus

With the development of digital technology, machine learning has paved the way for the next generation of tinnitus diagnoses. Although machine learning has been widely applied in EEG-based tinnitus analysis, most current models are dataset-specific. Each dataset may be limited to a specific range of symptoms, overall disease severity, and demographic attributes; further, dataset formats may differ, impacting model performance. This paper proposes a side-aware meta-learning for cross-dataset tinnitus diagnosis, which can effectively classify tinnitus in subjects of divergent ages and genders from different data collection processes. Owing to the superiority of meta-learning, our method does not rely on large-scale datasets like conventional deep learning models. Moreover, we design a subject-specific training process to assist the model in fitting the data pattern of different patients or healthy people. Our method achieves a high accuracy of 73.8% in the cross-dataset classification. We conduct an extensive analysis to show the effectiveness of side information of ears in enhancing model performance and side-aware meta-learning in improving the quality of the learned features.