Objective To address the diagnostic lag, low consistency, and subjective biases associated with traditional clinical scales (such as HAMD and DSM-5) and psychiatric interviews, this study aimed to establish objective, non-invasive electrophysiological biomarkers for major depressive disorder (MDD). By leveraging eyes-closed resting-state electroencephalography (rsEEG), which offers superior patient compliance and eliminates task-related cognitive confounding compared to task-state paradigms, we systematically investigated the spatial distribution patterns, disease-course-related non-linear evolution, and temporal robustness of the θ/β ratio (TBR) and θ/α ratio (TAR). Specifically, we aimed to map how these frequency-band ratios reflect neuropathological deficits in top-down cognitive control, cortical arousal, and emotional regulation across different severity tiers of depression, validating rsEEG as a reliable tool for clinical grading and screening.Methods A total of 154 right-handed participants (57 males, aged (41.02±16.93) years) were enrolled, following the exclusion of 12 subjects from an initial cohort of 166 due to environmental noise or poor electrode contact. Participants were stratified into healthy controls (HC, n=51), mild (n=40), moderate (n=21), and severe depression (n=42) groups based on HAMD, MMSE, and MoCA scales. After resting quietly for 30 min, eyes-closed rsEEG was recorded for 10 min using 19 electrodes positioned according to the international standard 10-20 system (sampling rate: 200 Hz; contact impedance<10 kΩ; bilateral mastoids A1/A2 reference) in a quiet, dimly lit room. Following offline filtering and manual artifact rejection of head movements, EMG, and ECG noises, a continuous, stable 3-min window was extracted for comparative analysis. Relative power spectral densities for the theta (4–7 Hz), alpha (8–12 Hz), and beta (13–30 Hz) bands were calculated to derive TBR and TAR across nine cortical regions (left/right anterior, middle, posterior regions, and midline Fz, Cz, Pz). Data were evaluated using 4 (groups)×9 (brain regions) two-way ANOVA with Bonferroni-adjusted post-hoc comparisons.Results Two-way ANOVA for TBR revealed highly significant main effects of group, brain region, and their interaction (all P<0.001). Abnormal TBR values were primarily observed in the frontal and temporal regions and exhibited significant disease-course-related changes. Specifically, TBR values were significantly higher in the mild and severe depression groups than in the healthy control and moderate depression groups (P<0.01), suggesting that depression-related EEG characteristics may undergo complex and dynamic changes as the disorder progresses, which may be modulated by clinical heterogeneity (such as the presence of comorbid anxiety and recurrent depression in the moderate group). TAR effectively differentiated depressive states and showed significant hemispheric asymmetry specifically prominent in the central (Cz, P<0.001) and posterior (Pz, P=0.019) regions, indicating a central-posterior regional bias in depression-related brain dysfunction. Stability analysis showed that TAR maintained stable between-group characteristics with only 3 min of EEG recording. In contrast, the healthy control""s TBR falsely exhibited a midline-dominant distribution in the 3-min segment, and its characteristic off-axis spatial asymmetry became stable only with a 10-min recording. These findings indicate that TAR has greater sensitivity detection and stronger temporal robustness than TBR.Conclusion TBR and TAR are sensitive indicators for assessing depression severity and associated neurophysiological alterations. Elevated TBR may indicate impaired cognitive control and reduced cortical arousal, whereas the spatial lateralization and nonlinear changes in TAR may reflect specific features of disease progression, potentially corresponding to interhemispheric information processing imbalance. Given its stability in recordings as short as 3 min, TAR may have considerable potential for efficient clinical assessment.