Dissociating tinnitus patients from healthy controls using resting-state cyclicity analysis and clustering

Abstract

Chronic tinnitus is a common and sometimes debilitating condition that lacks scientific consensus on physiological models of how the condition arises as well as any known cure. In this study, we applied a novel cyclicity analysis, which studies patterns of leader-follower relationships between two signals, to resting-state functional magnetic resonance imaging (rs-fMRI) data of brain regions acquired from subjects with and without tinnitus. Using the output from the cyclicity analysis, we were able to differentiate between these two groups with 58–67% accuracy by using a partial least squares discriminant analysis. Stability testing yielded a 70% classification accuracy for identifying individual subjects’ data across sessions 1 week apart. Additional analysis revealed that the pairs of brain regions that contributed most to the dissociation between tinnitus and controls were those connected to the amygdala. In the controls, there were consistent temporal patterns across frontal, parietal, and limbic regions and amygdalar activity, whereas in tinnitus subjects, this pattern was much more variable. Our findings demonstrate a proof-of-principle for the use of cyclicity analysis of rs-fMRI data to better understand functional brain connectivity and to use it as a tool for the differentiation of patients and controls who may differ on specific traits. Chronic tinnitus is a common, yet poorly understood, condition without a known cure. Understanding differences in the functioning of brains of tinnitus patients and controls may lead to better knowledge regarding the physiology of the condition and to subsequent treatments. There are many ways to characterize relationships between neural activity in different parts of the brain. Here, we apply a novel method, called cyclicity analysis, to functional MRI data obtained from tinnitus patients and controls over a period of wakeful rest. Cyclicity analysis lends itself to interpretation as analysis of temporal orderings between elements of time-series data; it is distinct from methods like periodicity analysis or time correlation analysis in that its theoretical underpinnings are invariant to changes in time scales of the generative process. In this proof-of-concept study, we use the feature generated from the cyclicity analysis of the fMRI data to investigate group level differences between tinnitus patients and controls. Our findings indicate that temporal ordering of regional brain activation is much more consistent in the control population than in tinnitus population. We also apply methods of classification from machine learning to differentiate between the two populations with moderate amount of success.

Keywords

• Tinnitus
• Resting-state fMRI
• Cyclicity
• Classification