Identifiability of Spectral Graph Model Parameters in Clinical MEG: Implications for Biophysical Interpretation and SOZ Localization

Abstract

Objective: We evaluate the practical identifiability and clinical utility of local spectral graph model(SGM) parameters estimated from resting-state magnetoencephalography (MEG) in drug-resistant epilepsy. Methods:A coupled excitatory-inhibitory SGM was fitted to MEGpower spectra across 159 brain regions in 20 patients withtemporal lobe epilepsy who achieved seizure freedom following surgery. Identifiability was assessed via boundarysaturation analysis and inter-parameter correlations acrossfour frequency bands. Identifiable parameters were testedfor seizure onset zone (SOZ) discrimination. Results: Modelfit was excellent (mean r = 0.976) and significantly exceeded a 1/fβ baseline (p < 10−10). Gain parameters (gei,gii) were robustly estimable, whereas the excitatory timeconstant (τe) showed 74% boundary saturation in broadband fits, reduced to ∼3% when restricted to 1–50 Hz. SOZregions exhibited elevated gei (Cohen’s d = +0.67, FDRcorrected p = 0.024) and reduced gii (d = −0.55, p = 0.003),with a composite biomarker achieving 2.6-fold improvement over chance. Conclusion: Gain parameters are robustly identifiable from clinical MEG and capture excitatoryinhibitory imbalance in the SOZ, whereas time constantsrequire band-limited fitting. These findings motivate anidentifiability-aware framework in which only parametersdemonstrably constrained by data are interpreted. Significance: This is the first systematic assessment of neuralmass model parameter identifiability in clinical epilepsyMEG, establishing practical guidelines for biophysical parameter interpretation. © 1964-2012 IEEE.