Introduction Drug-resistant focal epilepsy often requires identifying the seizure onset zone (SOZ) for resection or neuromodulation, yet objective biomarkers of SOZ excitability remain limited. Neural adaptation across multiple timescales can reshape EEG power spectra and can be parameterized using fractional-order dynamics. We hypothesize that the fractional order (alpha) of a fractional neuronal network model provides an indirect, mechanistically grounded measure of neuronal excitability that can be inferred from macroscopic recordings, with potential utility for SOZ localization [2-4].
Methods We generated synthetic datasets by simulating a recurrent fractional-order neuronal network in which each neuron included a fractional-order filter implementing fractional differentiation consistent with cortical pyramidal-neuron adaptation [1,2]. Networks were driven by white-noise current input; alpha was varied while all other parameters were fixed. From inputs and network outputs we extracted phase shift, phase-locking value, power spectral density (PSD) slope and band powers, spectral density, and Hilbert-spectrum metrics. We fit regression models relating each feature to α and compared goodness of fit across features.
Results Across different α values, the PSD slope of the network output showed the clearest and most consistent relationship with α. This trend was roughly monotonic. In contrast, phase-based features and Hilbert-spectrum measures were weaker and more variable. Since a fractional differentiator changes the spectrum as a function of frequency, the changes in PSD slope provide an interpretable link between signal properties and alpha. These results suggest that PSD slope could be a simple surrogate marker for fractional order and, indirectly, for neuronal excitability in EEG-like signals.
Discussion Spectral slope has been linked to synaptic excitation/inhibition balance and hyperexcitability, and interictal EEG near the SOZ shows flattened PSD slopes consistent with reduced adaptation and increased excitability. Our simulation study suggests that estimating alpha from PSD slope could provide a mechanistically grounded, low-complexity biomarker for SOZ identification. Next, we will apply the α-estimation pipeline to clinical EEG recordings with electrode-level SOZ annotations, evaluating SOZ vs non-SOZ classification performance and robustness across patients and recording states.
References
1. Lundstrom, B. N., et al. (2008). Fractional differentiation by neocortical pyramidal neurons. Nat Neurosci. https://doi.org/10.1038/nn.2212 2. Lundstrom, B. N., & Richner, T. J. (2023). Neural adaptation and fractional dynamics as a window to underlying neural excitability. PLoS Comput Biol. https://doi.org/10.1371/journal.pcbi.1010527 3. Gao, R., et al. (2017). Inferring synaptic excitation/inhibition balance from field potentials. NeuroImage. https://doi.org/10.1016/j.neuroimage.2017.06.078 4. Lundstrom, B. N., et al. (2021). Low frequency novel interictal EEG biomarker for localizing seizures and predicting outcomes. Brain Commun. https://doi.org/10.1093/braincomms/fcab231
Acknowledgement No funding was received for this work.
