Deep learning-based electroencephalic diagnosis of tinnitus symptom

Area of Science:

• Neuroscience
• Medical Diagnostics
• Artificial Intelligence

Background:

• Tinnitus diagnosis relies on subjective and complex methods.
• Objective diagnostic tools for tinnitus are needed.

Purpose of the Study:

• To develop a deep learning model for tinnitus diagnosis using electroencephalographic (EEG) signals.
• To identify neural signatures of tinnitus during auditory cognitive tasks.

Main Methods:

• Utilized the EEGNet deep learning model to analyze EEG signals from patients performing active and passive oddball tasks.
• Analyzed EEGNet convolutional kernel feature maps and performed time-frequency analysis.

Main Results:

• The EEGNet model achieved an area under the curve of 0.886 in identifying tinnitus patients during an active oddball task.
• Significantly reduced pre-stimulus alpha activity was observed in the tinnitus group compared to the healthy group.
• Differences in evoked theta activity were noted during target stimuli in the active oddball task.

Conclusions:

• Task-relevant EEG features, particularly alpha and theta activity, serve as potential neural signatures for tinnitus.
• An EEG-based deep learning approach shows feasibility for objective tinnitus diagnosis.