Spectral-spatial Modeling of Neural Couplings in Functional Connectivity of Motor Networks

Functional neural connectivity reflects coordinated interactions within brain networks in both health and disease. Phase-amplitude coupling (PAC) is widely used for this purpose, but its pairwise formulations overlook spatially distributed, spectrally correlated structures in neural activity that arises from complex brain circuits.

We developed a spectral-spatial generalized linear model (SS-GLM) to characterize the high-frequency response as the integrated outcome of multiple low-frequency phase sources across channels and frequencies. Our method also utilizes a regularization strategy to mitigate predictor collinearity and spurious effects for stable attribution of network-level coupling estimates.

Validation was performed using both synthetic data and human motor network recordings from 25 patients with Parkinson's disease undergoing deep brain stimulation (DBS) surgery. In synthetic analyses, SS-GLM accurately recovered ground-truth coupling configurations under multiple plausible scenarios. In real data, the framework revealed coherent cortico-subthalamic coupling patterns across distinct motor tasks that aligned with clinically effective DBS contacts and their anatomical organization.

Conventional pairwise PAC metrics, such as the modulation index and GLM assess interactions in isolation. In contrast, SS-GLM jointly models distributed predictors and yields spatially localized, spectrally specific coupling estimates that also achieve stronger correspondence with clinical stimulation outcomes.

This work offers a principled and extensible framework for advancing functional connectivity analyses, with improved interpretability of network-level interactions, enhanced robustness to confounds, and broad potential for future applications in systems neuroscience and neuro-engineering research.