How Artificial Intelligence and Brain Connectivity Can Inform Outcome Prediction in Drug-Resistant Epilepsy

Epilepsy affects over 50 million people worldwide, and about one-third of them develop drug-resistant epilepsy (DRE), where seizures persist despite antiseizure medications. Therefore, early identification of drug-resistant patients is fundamental to ensuring timely and appropriate treatment planning. Surgical resection of the epileptogenic zone (EZ) remains the most effective option for treating patients with DRE, yet almost 40% of the patients continue to experience seizures postoperatively. This highlights the need for a more accurate preoperative localization of the EZ. For patients who are not candidates for resection, vagus nerve stimulation (VNS) is a possible therapeutic option that provides palliative benefits by reducing seizure frequency. This scoping review examines how structural, functional, and effective connectivity—derived from neurophysiological and neuroimaging brain data—has been combined with artificial intelligence (AI) to address key clinical questions: predicting drug resistance, localizing the EZ, predicting postsurgical outcomes, and estimating VNS treatment response. Forty-one studies demonstrated that differences in structural and functional connectivity patterns can distinguish refractoriness and predict treatment outcomes. Favorable postsurgical results were consistently associated with the resection of highly connected or aberrant network hubs that overlapped with the EZ. Integrating connectivity measures with AI approaches frequently improved classification compared to other well-established biomarkers, underscoring their value as complementary tools in presurgical evaluation and treatment planning. Despite methodological heterogeneity, small sample sizes, and limited explainability, these results outline a future of more individualized epilepsy care.