Seizure Forecasting in the Intra-hippocampal Kainic Acid Mouse Model
Abstract number :
2.086
Submission category :
3. Neurophysiology / 3F. Animal Studies
Year :
2023
Submission ID :
1059
Source :
www.aesnet.org
Presentation date :
12/3/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Sam McKenzie, PhD – UNM
Davey Gregg, BSc – UNM; Christos Lisgaras, PhD – NYU Nathan Kline Institute; Helen Scharfman, PhD – NYU Nathan Kline Institute; Alexandra Sommer, Masters – UNM
Rationale: Responsive neural stimulation (RNS) is effective in controlling seizures for those with medication resistant epilepsy. The mechanism of action underlying the therapeutic value of RNS is not well understood, as the overall patient outcome correlates poorly with acute seizure termination. Efficacy also significantly improves over time, suggesting that RNS induces beneficial neural plasticity. To gain a better understanding of how RNS treats focal epilepsy, and to develop responsive algorithms that are anticipatory rather than reactive, pre-clinical focal epilepsy models are needed.
Methods: The intra-hippocampal kainic acid (IHKA) mouse model produces subjects that experience spontaneous convulsions. Recently, a large database (N = 5 mice; 413 seizures, 1512 hrs) of daily 24-hour continuous video/EEG recordings was reported (Lisgaras and Scharfman, 2022). This database was used to identify a seizure vulnerable state and to develop algorithms to forecast spontaneous seizures in this mouse model. The local field potentials were taken from four skull screws and the state space at each moment in time (1s window) was defined by >300 features including: the wavelet power spectra, phase-amplitude coupling, and cross-channel coherence.
Results: We observed several changes in the hour prior to a seizure: overall delta power increased, hippocampal and cortical low frequency coherence decreased (< 5Hz), and delta/gamma phase amplitude coupling increased. In the minutes before a seizure, gamma power was low while delta power was high. Visual inspection of the probability density function in a 2-D state space revealed that seizures arise from a narrow range of brain states, however, this state space is also represented during moments when seizures are not imminent. This combination of results suggests that a classifier should show low false positives (high specificity) and high false negatives (low sensitivity) in detecting whether a seizure will occur in the near future. Because seizures are rare, we used the RUSBoost algorithm to build a set of weak classifiers to predict discretized time until a seizure. Indeed, we found high specificity ( >
Neurophysiology