Frequency-Specific Resting-State MEG Network Characteristics of Tinnitus Patients Revealed by Graph Learning

Tinnitus, the perception of sound without an external source, affects a significant portion of the population, yet its impact on brain communication diagram known as the functional connectome, remains limited. Traditional functional connectivity (FC) methods, such as Pearson correlation, phase lag index and coherence rely on pairwise comparisons and are therefore limited in providing a holistic encoding of FC. Here, we employ an alternative approach to estimate the entire connectivity structure by analyzing all time-courses simultaneously. This approach is robust even for short-duration recordings, facilitating faster functional connectome identification and real-time applications. Using resting-state MEG recordings from controls and individuals with tinnitus, we demonstrated that the learned connectomes outperform correlation-based connectomes in fingerprinting, that is, identifying an individual from test-retest acquisitions. Group-level analysis revealed distinct altered FC in tinnitus across multiple frequency bands, affecting the default mode, auditory, visual, and salience networks, suggesting a reorganization of these large-scale networks beyond auditory areas. Our study reveals that tinnitus presents highly individualized and heterogeneous whole-brain connectome profiles, highlighting the need to focus on individual variability rather than group-level differences to gain a more nuanced understanding of tinnitus. Personalized FC could enable patient-specific tinnitus models, optimizing treatment strategies for individualized care.