Abstracts

Rationale: Epilepsy and seizures are associated with changes in sleep. The transition into a seizure, which includes an impairment of consciousness, can be seen as a shift in the arousal state. EEG spectral measures have been linked to arousal changes. Slow wave activity (SWA, 0.5-4Hz) is tied to sleep depth and inversely related to environmental responsiveness. Frontal theta (4-8Hz) correlates with sleep pressure, while posterior alpha (8-12Hz) is linked to quiet wakefulness. Sigma power (10-15Hz) is associated with spindle activity in sleep stage 2. However, the interaction between these characteristics is not fully understood. Here, we demonstrate how to combine these measures and create a simple low-dimensional state-space consistent across subjects, providing a framework to visualize and analyze physiological and pathological events like sleep-wake transitions and seizure dynamics.


Methods: We utilized the EPILEPSIAE database, which contains EEG data collected during epilepsy monitoring unit admissions. Spectral profiling of the EEG was performed on consecutive time windows across 25,000 hours of EEG data from 158 patients. To identify feature combinations capturing most of the variability in the signal, we used data from 118 patients to learn a principal component analysis (PCA) transformation. We then tested the applicability of this transformation on the remaining 40 subjects, which were not used for training. To map the sleep stages in this space, we used the YASA sleep scoring system, which provides an output of the sleep stage and its probability. Lastly, we analyzed the seizure trajectory through the pre-ictal, seizure onset, offset, and postictal points in the new state space. For each principal component (PC), we analyzed seizure trajectory in the context of sleep state in the pre-ictal epoch, seizure localization (frontal-temporal), and seizure clinical classification.


Results: The first PC accounted for 58% of the variability, while the first three components accounted for 75%. A clear separation between wakefulness and non-rapid eye movement sleep stages 2-3(Figure 1A) was present in the three-dimensional component space. Temporal analysis of the first principal components revealed robust representations of circadian and ultradian cycles(Figure 1B). Exploration of the seizure state within this space showed that seizures are consistently found on the outskirts of the waking state. Analysis of seizure trajectories in PC1 converged into a state that is close to wakefulness regardless of the pre-ictal state. Comparing frontal to temporal lobe seizures revealed a higher increase in arousal for frontal seizures. Lastly, the generalization of seizures at offset was evident, with a prominent increase in arousal at seizure offset and a sharp postictal decrease(Figure 1C).


Conclusions: Our representation of the EEG spectral profile introduces a novel approach for analyzing global brain states and seizure dynamics across different time scales. This method provides insights into the interactions between arousal states and seizure activity, potentially informing clinical approaches to epilepsy management.


Funding: This research was supported by Israel Science Foundation grant 794/22 to O.S.