Exploring the Relationship Between Multiday Cycles in Intracranial EEG Features and Cycles in Seizure Clusters
Abstract number :
3.442
Submission category :
2. Translational Research / 2E. Other
Year :
2023
Submission ID :
1427
Source :
www.aesnet.org
Presentation date :
12/4/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Krishnakant Saboo, PhD – University of California San Francisco
Yurui Cao, BS – University of Illinois Urbana Champaign; Vaclav Kremen, PhD, MS – Mayo Clinic; Vladimir Sladky, BS – Mayo Clinic; Nicholas Gregg, MD – Mayo Clinic; Paul Arnold, MD, FACS – Carle Foundation Hospital; Suguna Pappu, MD, PhD – Carle Foundation Hospital; Philippa Karoly, PhD – University of Melbourne; Dean Freestone, PhD – Seer Medical; Mark Cook, MD – University of Melbourne; Ravishankar Iyer, PhD – University of Illinois Urbana Champaign; Gregory Worrell, MD, PhD – Mayo Clinic
Rationale: Few studies have explored whether cycles manifest differently for isolated and clustered seizures and their relationship to cycles in inter-ictal intracranial EEG (iEEG) features. Such an investigation can be valuable in forecasting seizures using characteristics of iEEG features. Additionally, seizure consolidation has been hypothesized to play a role in epilepsy progression. Thus, within a data-driven model, we studied whether seizure consolidation modified seizure cycles. Finally, we investigated statistical models that combine long-term iEEG features and consolidation to track and forecast seizures.
Methods: We used several months-long chronic iEEG recordings of nine people with epilepsy [1]. Seizures within 24 hours of each other were considered part of a cluster. To test for cycles in seizures, we computed R-values for cycle periods of 2 to 128 days for isolated and clustered seizures separately. We extracted a univariate (power in band) and several bivariate iEEG features (relative entropy, phase locking value, correlation coefficient) to represent the iEEG activity of each day. To study iEEG cycles, we looked for significant peaks in the wavelet periodogram of each iEEG feature time series. Finally, we developed a patient-specific state space model (SSM) with the unobserved disease state as the latent variable, seizures as control input, and iEEG features as observations. The model assumed that cycles in the latent disease state led to cycles in seizures and iEEG features, and consolidation influenced the latent state. The SSM was trained using iEEG features to estimate the latent disease states.
Results: Multiday cycles were observed in seizure clusters in eight out of nine patients and in isolated seizures in two out of nine patients. Cycle periods were patient-specific and ranged from 5 to >50 days. For Patient 9 (Pt9), the most prominent cycle was 50 days long. Cycles in iEEG features were studied only for Pt9 because of missing data in feature time series of other patients. Multiday cycles were observed across univariate and bivariate features, electrodes, and physiological frequency bands with several features having cycle periods among 3.5, 6.9, 49.5, and 124.6 days. Since multiday cycles were observed in iEEG features and seizure clusters, an SSM was trained for Pt9 using select iEEG features. five out of nine clusters of Pt9 occurred during peaks in the cycles of latent disease state. Thus, the SSM trained with long-term iEEG features might predict periods of heightened risk of seizure clusters. The control input matrix, which captured the effect of seizure consolidation on the latent disease state, had non-zero coefficients, suggesting that seizures with varying spectral signatures had varying effects on the cycles.
Conclusions: Our SSM statistically related multiday cycles in iEEG features and seizure clusters, as well as consolidation. The model can potentially forecast clusters days in advance, but prospective studies are needed.
Reference: [1] Cook, Lancet Neurol, 2013.
Funding: NSF CNS-1624790, NIH UH2&3 NS095495, NIH R01 NS092882, Mayo/Illinois Alliance Fellowship.
Translational Research