Resting-State EEG Biomarkers of Tinnitus Robust to Cross-Subject and Cross-Platform Variation

A team led by Adyant Balaji and Gert Cauwenberghs at UC San Diego has developed a novel EEG-based biomarker for tinnitus that generalizes across different subjects and recording platforms. The study, published on arXiv, applied Koopman operator analysis via Dynamic Mode Decomposition (DMD) to resting-state EEG data from two independent datasets. This approach extracts features related to the stability and frequency of neural oscillations, outperforming traditional microstate analysis in classifying tinnitus versus control subjects.

The key finding is that the magnitude of Koopman eigenvalues—reflecting oscillatory decay rates—shows strong cross-dataset consistency (mean correlation ρ = 0.685), while the phase (oscillation frequency) does not (ρ = 1.583). This indicates that altered oscillatory stability, not frequency shifts, is the more robust tinnitus biomarker. The linear SVM classifier using PCA-reduced Koopman features achieved the best cross-dataset generalization, a critical step toward clinical adoption for objective tinnitus diagnosis.