Structural brain imaging biomarkers for predicting seizure recurrence after a first unprovoked seizure.
Authors
Affiliations (7)
Affiliations (7)
- The Florey Institute of Neuroscience and Mental Health, Melbourne Brain Centre, Melbourne, Victoria, Australia.
- Florey Department of Neuroscience and Mental Health, The University of Melbourne, Melbourne, Victoria, Australia.
- Department of Neurology, Austin Health, Heidelberg, Victoria, Australia.
- Department of Clinical Neuropsychology, Austin Health, Heidelberg, Victoria, Australia.
- Eastern Health Clinical School, Monash University, Box Hill, Victoria, Australia.
- Department of Neurology, Northern Health, Epping, Victoria, Australia.
- Faculty of Medicine, Dentistry and Health Sciences, The University of Melbourne, Melbourne, Victoria, Australia.
Abstract
Predicting seizure recurrence following a first unprovoked seizure (FUS) remains a significant clinical challenge, especially when routine clinical magnetic resonance imaging (MRI) and EEG do not reveal abnormalities diagnostic of epilepsy. Here, we incorporate quantitative structural MRI-derived biomarkers into prediction models for seizure recurrence at 12 months and identify brain structural features that are predictive of seizure recurrence. We analyzed a retrospective, multicenter cohort of 197 adult patients with FUS, comprising 83 with seizure recurrence and 114 with no seizure recurrence at 12 months. All participants had normal or nondiagnostic MRI and EEG findings. Morphometric features were extracted from clinical 3 T T1-weighted MRI using FreeSurfer. Machine learning algorithms were trained on combined imaging and clinical features using nested cross-validation for model selection. Performance was compared with a logistic regression model based on clinical features only. The best-performing model, a support vector machine (SVM) trained on a combination of imaging features and clinical factors, achieved an AUC of 0.65 (95% CI: 0.57-0.73), significantly better than chance (p = 0.01 when compared with an AUC of 0.5). In contrast, the logistic regression model trained on clinical factors alone yielded an AUC of 0.57 (95% CI: 0.49-0.65), not statistically different to chance (p = 0.28). Direct comparison between the SVM and the logistic regression clinical factor-only model was not statistically significant (95% CI for the difference in AUC: -0.019 to 0.173, p = 0.11). The most important imaging features for prediction were inter-hemispheric asymmetry of subcortical and cortical gray matter volumes and regional gyral curvatures, particularly in fronto-parietal and limbic regions. Quantitative structural MRI contributes additional information beyond clinical factors for machine learning models predicting seizure recurrence. Changes to cortical folding and gray matter asymmetries in cortical and subcortical regions show potential as prognostic biomarkers of seizure recurrence risk after a FUS. Identifying individuals who will have another seizure after their first unprovoked seizure is difficult when routine brain scans and EEG appear normal. We developed a tool that combines MRI-derived markers with clinical information to predict seizure recurrence. Subtle structural differences in the brain, especially asymmetries between left and right hemispheres and changes to cortical folding, were associated with a higher chance of another seizure within a year. This approach has potential in identifying individuals at risk of seizure recurrence earlier.