목적: Nonpulsatile tinnitus, a perception of sound in the absence of an external sound source, is a purely subjective symptom as it can only be observed by the person who suffer from tinnitus. It would be highly desirable to diagnose the presence or absence of tinnitus in an objective way. Recently, scientists have developed machine learning techniques that can learn to recognize patterns by classifying seen data, taking into account their statistical variation. These algorithms can subsequently be applied to unseen data. In other words, based on the known properties learned from the trained data, these algorithms can predict whether the pattern corresponds to the presence or absence of tinnitus. We therefore combined resting-state quantitative electroencephalography (rs- qEEG) with machine learning to develop a brainbased electrophysiological signature for the presence or absence of tinnitus. 방법:One hundred and fiftythree tinnitus patients and 264 healthy controls underwent rs-qEEG measurements for 5 minutes. These data were used as training sets, and the predictability of the presence of tinnitus was trained using a support vector machine. Regions of interest were the auditory cortex, dorsal- and subgenual anterior cingulate cortex, posterior cingulate cortex, parahippocampus, and insula. 결과:Using the support vector machine, the current yielded better predictive results than using Bayesian inference learning, with a correct predictability of approximately 90%. In other words, presence of tinnitus could be predicted only by rs-qEEG findings with a correct predictability of 90%. 결론:Taken together, the current study suggests that it might become possible to diagnose the presence or absence of tinnitus based solely on an EEG oscillatory signature in the near future.