Graph-theory based degree centrality combined with machine learning algorithms can predict response to treatment with antiepileptic medications in children with epilepsy.

Background And Purpose: The purpose of the current study is to detect changes of graph-theory-based degree centrality (DC) and their relationship with the clinical treatment effects of anti-epileptic drugs (AEDs) for patients with childhood absence epilepsy (CAE) using resting-state functional MRI (RS-fMRI).

Methods: RS-fMRI data from 35 CAE patients were collected and compared with findings from 35 age and gender matched healthy controls (HCs). The patients were treated with AEDs for 46.03 weeks before undergoing a second RS-fMRI scan.

Results: CAE children at baseline showed increased DC in thalamus, postcentral and precentral and reduced DC in medial frontal cortex, superior frontal cortex, middle temporal cortex, angular and precuneus. However, those abnormalities showed a clear renormalization after AEDs treatments. We then explored the viability of graph-theory-based degree centrality to accurately classify effectiveness to AEDs. Support Vector Machine analysis using leave-one-out cross-validation achieved a correct classification rate of 84.22% [sensitivity 78.76%, specificity 89.65%, and area under the receiver operating characteristic curve (AUC) 0.96] for differentiating effective subjects from ineffective subjects. Brain areas that contributed most to the classification model were mainly located within the right thalamus, bilateral middle temporal gyrus, right medial frontal gyrus, right inferior frontal gyrus, left precuneus, bilateral angular right precentral and left postcentral. Furthermore, the DC change within the bilateral angular are positively correlated with the symptom improvements after AEDs treatment.

Conclusion: These findings suggest that graph-theory-based measures, such as DC, combined with machine-learning algorithms, can provide crucial insights into pathophysiological mechanisms and the effectiveness of AEDs.