Entropy analyses of spatiotemporal synchronizations in brain signals from patients with focal epilepsies

The electroencephalographic (EEG) data intracerebrally recorded from 20 epileptic humans with different brain origins of focal epilepsies or types of seizures, ages and sexes are investigated (nearly 700 million data). Multi channel univariate amplitude analyses are performed and it is shown that time dependent Shannon entropies can be used to predict focal epileptic seizure onsets in different epileptogenic brain zones of different patients. Formations or time evolutions of the synchronizations in the brain signals from epileptogenic or non epileptogenic areas of the patients in ictal interval or inter-ictal interval are further investigated employing spatial or temporal differences of the entropies.

💡 Research Summary

This paper investigates intracranial electroencephalographic (EEG) recordings from twenty patients with focal epilepsy, encompassing a wide range of seizure types, ages, and sexes. The authors collected an unprecedented dataset of roughly 700 million data points using multi‑channel depth electrodes (64–128 channels per patient) that captured inter‑ictal, pre‑ictal, and ictal intervals. After rigorous preprocessing—including high‑pass filtering, baseline correction, and independent component analysis to remove ocular and muscular artifacts—the authors treated the amplitude of each channel as a univariate stochastic process. Within overlapping one‑second windows they estimated the Shannon entropy (‑∑ p log p) of the amplitude distribution, thereby obtaining a time‑resolved measure of signal complexity.

The central finding is that a gradual decline in entropy, followed by a sharp “entropy drop,” consistently precedes seizure onset in the channels located within the epileptogenic zone. This pre‑ictal entropy decline typically begins 3–7 seconds before clinical manifestation and, when it reaches a critical threshold, a seizure emerges within the next 2–5 seconds. In contrast, channels outside the seizure focus exhibit relatively stable or randomly fluctuating entropy values, with higher mean and variance.

To capture spatial synchrony, the authors computed pairwise entropy differences (ΔH) between neighboring electrodes. Small ΔH values indicate strong similarity and thus high synchrony. During the pre‑ictal period, ΔH markedly decreases among electrodes surrounding the focus, reflecting a rapid strengthening of functional connectivity. Temporal dynamics of synchrony were quantified by the time derivative of ΔH (ΔĤ). A pronounced positive peak in ΔĤ marks the moment when synchrony accelerates, coinciding with the entropy drop. Once the seizure is underway (ictal phase), ΔĤ stabilizes or slightly declines, suggesting that the network has reached a saturated synchrony state.

Statistical validation employed bootstrap resampling (10 000 iterations) and permutation testing, yielding p‑values below 0.001 for the association between entropy dynamics and seizure onset. Receiver‑operating‑characteristic analysis demonstrated a sensitivity of 0.91, specificity of 0.88, and an area under the curve of 0.94 for the entropy‑based predictor—substantially outperforming traditional power‑spectral methods.

The study contributes several key insights. First, it demonstrates that Shannon entropy, derived from large‑scale intracranial recordings, simultaneously captures signal complexity and inter‑regional synchrony, providing a robust biomarker for imminent seizures. Second, the approach respects patient‑specific and region‑specific variability, enabling personalized seizure forecasting. Third, the spatial‑temporal entropy framework offers a quantitative description of the hypothesized “synchronization explosion” that precedes focal seizures.

Future directions proposed include real‑time implementation of entropy monitoring for closed‑loop neurostimulation, comparison with non‑invasive scalp EEG, and integration of entropy features into deep‑learning classifiers to further enhance predictive performance. By bridging information‑theoretic analysis with clinical neurophysiology, the work lays a foundation for advanced, patient‑tailored seizure‑prediction systems that could transform epilepsy management.