Określanie strony ogniska padaczki za pomocą uczenia maszynowego i analizy białej substancji mózgu

PubMed➕ 20.11.2026medRxiv

Epilepsy Lateralization by Elastic Net Regularization of a Diffusion Tensor Imaging Biomarker using Machine Learning

W skrócie

Badacze opracowali komputerowy model, który analizuje specjalne obrazy mózgu (DTI) oraz dane kliniczne pacjentów, aby dokładnie określić, z której strony mózgu pochodzi padaczka oporná na leki. Model działał dobrze, szczególnie w rozróżnianiu padaczki lewostronnej od padaczki obustronnej lub uogólnionej, osiągając dokładność 90%. Wyniki sugerują, że ta metoda może w przyszłości wspomóc lekarzy w planowaniu operacji padaczki, choć potrzeba jeszcze dalszych badań przed wprowadzeniem jej do praktyki klinicznej.

Oryginalny abstract (angielski)

OBJECTIVE: Accurate lateralization of the epileptogenic focus is essential for presurgical evaluation of patients with drug-resistant epilepsy. This study examined whether diffusion tensor imaging (DTI) measures of white matter microstructure, combined with clinical and demographic variables, could classify epilepsy lateralization using machine learning. METHODS: We present a retrospective analysis of 334 adult patients undergoing presurgical epilepsy evaluation at the University of Kansas Health System. Fractional anisotropy (FA) and mean diffusivity (MD) were extracted from 36 bilateral white matter tracts and combined with clinical and demographic variables. Several supervised machine learning approaches were evaluated, with elastic net logistic regression selected for detailed analysis. Models classified right-sided, left-sided, and bilateral or generalized epilepsy. Performance was evaluated with area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity as performance metrics. RESULTS: Elastic net models demonstrated good discrimination across classification tasks, with AUC values ranging from 0.76 to 0.88. The strongest performance distinguished left-sided epilepsy from bilateral or generalized epilepsy (AUC = 0.88; accuracy = 0.90; sensitivity = 0.90; specificity = 0.89). Discrimination between unilateral epilepsy and bilateral or generalized epilepsy was also good (AUC = 0.83). Classification of right-sided versus left-sided epilepsy demonstrated lower performance (AUC = 0.76). SIGNIFICANCE: DTI-derived white matter features analyzed using a machine learning approach may provide complementary information for epilepsy lateralization during presurgical evaluation. Further studies incorporating other pre-surgical evaluation data into the machine learning model are needed to establish the clinical utility of this approach. KEY POINTS: A machine learning model was tested using diffusion tensor imaging and clinical data to retrospectively classify epilepsy lateralization.Our model balanced sensitivity and specificity in epilepsy localization, suggesting potential as an interpretable surgical decision tool.Findings support the hypothesis that white matter tracts contain useful information about seizure propagation to distal brain regions.Further model validation is needed before this can be used as a clinical tool.

Metadane publikacji

Journal
medRxiv
Data publikacji
04.11.2026
PMID
43060756
DOI
10.64898/2026.11.03.26367189
Autorzy
Lepping RJ, Pearl M, Chen X, Hu J, Li Y, Akhter M, Farrenburg M, Landazuri P, Pearson CM
Źródło
PubMed