Wieloskalowa analiza regularności oscylacji gamma umożliwia precyzyjne przewidywanie napadów padaczki i lokalizację strefy epileptogennej
Dual-Time Scale Surveillance of Gamma Oscillation Regularity Dynamics Enables High- Precision Seizure Prediction and Epileptogenic Zone Localization
W skrócie
[Preprint - wstępne wyniki] Badacze opracowali nową metodę analizy aktywności elektrycznej mózgu, która łączy dwa podejścia: jedno do znalezienia miejsca pochodzenia napadów (z dokładnością 68,8% dla jednej strefy) i drugie do przewidzenia ich występowania (z dokładnością 81,2% na 5 sekund przed napadem). Połączenie obu metod pozwoliło na prawidłowe przewidzenie napadów i wskazanie ich źródła u 93,8% pacjentów (15 na 16), co stanowi obiecujący krok w kierunku lepszego leczenia padaczki poprzez stymulację mózgu.
Oryginalny abstract (angielski)
Abstract Objective: Accurate spatial localization of the epileptogenic zone (EZ) and temporal prediction of seizure onset are the two most critical hurdles in epileptology. Building upon previous pathophysiological findings of preictal inhibitory breakdown, we aimed to establish the optimal temporal resolution for gamma oscillation regularity (GOR) dynamic markers to capture the preictal phase transition, proposing a highly accurate hybrid architecture for both localization and prediction. Methods: We evaluated 16 patients with focal epilepsy from the CHB-MIT scalp EEG dataset. Data were downsampled to 200 Hz and filtered within the gamma band (30–80 Hz). To assess spatial and temporal dynamics, we computed the static interictal GOR using multiscale entropy and the dynamic preictal information flow surge using the temporal difference of net transfer entropy (ΔGOR-TE surge). Analytical time windows were swept across 5, 10, 15, and 20 seconds. Additionally, a continuous sliding-window analysis over the preictal 60 seconds was performed on a rigorously defined sub-cohort of 13 patients with artifact-free continuous data to validate prediction specificity. Results: The static interictal GOR achieved its maximal lobar concordance (68.8%) using a 10-second rolling window for precise EZ localization. In contrast, the dynamic ΔGOR-TE surge peaked (81.2%) exclusively within a 5-second window immediately preceding clinical onset. Continuous preictal tracking in the sub-cohort demonstrated a highly stable physiological baseline characterized by a strict absence of premature predictive alerts, culminating in an abrupt, near-vertical information surge exactly 5 seconds before onset. By integrating these temporally optimized markers, the logical union successfully localized the epileptogenic lobe and predicted the impending seizure with 93.8% accuracy (15/16 patients). Significance: Seizure onset is a catastrophic non-linear phase transition. Our findings demonstrate that combining a 10-second interictal GOR with a 5-second ΔGOR-TE preictal surveillance provides an exceptionally robust, dual-time-scale architecture for precise seizure prediction and next-generation responsive neurostimulation.