Identyfikacja cech zapisów mózgu (EEG) podczas stanów padaczkowych do przewidywania rodzaju epilepsji u myszy na modelu kwasu kaninowego wstrzykiwanego do migdałka mózgu
Identification of EEG features during status epilepticus for prediction of emergent epilepsy phenotype in the mouse intra-amygdala kainic acid model using supervised learning
W skrócie
Naukowcy badali, czy zapisy aktywności elektrycznej mózgu wykonane na początku ataku padaczkowego mogą przewidzieć, ile napadów padaczkowych będzie miał pacjent później. Używając sztucznej inteligencji do analizy zapisów z mózgu myszy, wykazali, że mogą rozróżnić zwierzęta, które będą miały mało, średnio lub dużo napadów w przyszłości. Badanie pokazuje, że wczesna aktywność elektryczna mózgu zawiera ważne informacje o tym, jak ciężka będzie epilepsja, co może pomóc w przyszłości w przewidywaniu przebiegu choroby i doborze leczenia.
Oryginalny abstract (angielski)
Preclinical animal models are essential for investigating epilepsy mechanisms and evaluating novel therapies. In rodents, epilepsy can be induced by status epilepticus, leading to later spontaneous recurrent seizures (SRSs). However, inter-animal variability in seizure burden can limit suitability for drug studies. Here, we investigated whether early electroencephalography (EEG) recordings acquired during status epilepticus (∼40 min) can predict the later burden of spontaneous seizures in the intra-amygdala kainic acid (IAKA) model in mice. Spectral and statistical EEG features were extracted from IAKA model mice (n = 19) and used to train supervised classifiers, including Random Forest, Support Vector Machine (SVM), and Logistic Regression. Performance was evaluated using Leave-One-Out and 5-fold-stratified cross-validation. Feature robustness was enhanced through an intersection-based strategy combining ANOVA, Mutual Information, Random Forest importance, and SHAP analysis, together with engineered change-based (∆) features derived as post-KA minus baseline activity. Among the evaluated models, SVM achieved the strongest internal performance (weighted F1-score: 0.74), demonstrating that early EEG dynamics during status epilepticus encode prognostic information related to later SRSs burden. Importantly, the model reliably identified animals within the intermediate (normal) seizure-burden group that showed an average of two-week seizure count of 53 ± 11, relative to low (18 ± 11) and high (85 ± 43) groups. While generalization to an independent validation cohort (n = 11) was limited, this likely reflects biological heterogeneity and duration-dependent EEG variability. Overall, these findings highlight the relevance of early electrographic activity in shaping seizure-burden outcomes and provide a foundation for future longitudinal prognostic studies.