Artificial intelligence for adaptive neuromodulation in drug-resistant epilepsy

Drug-resistant epilepsy (DRE) affects nearly one third of people with epilepsy and is associated with substantial cognitive, psychiatric, and mortality burdens. For patients who are not candidates for resection or laser interstitial thermal therapy, neuromodulation therapies such as vagus nerve stimulation, deep brain stimulation, and responsive neurostimulation offer an important and established therapeutic option that can provide substantial, often life-changing benefits, although clinical response varies widely and typically requires prolonged, iterative optimization. Epilepsy is increasingly understood as a disorder of distributed and dynamic brain networks, in which seizure generation, propagation, and termination reflect interactions among connected regions that evolve over time. In this setting, improving neuromodulation requires precise and patient-specific decisions across the treatment workflow. This perspective reframes neuromodulation as a control problem organized around four interdependent clinical questions: who is most likely to benefit, where to deliver stimulation within an individual's epileptic network, when to apply stimulation relative to evolving seizure risk, and how to configure and adapt stimulation parameters over time. Artificial intelligence (AI) may provide a unifying framework for addressing these questions in a personalized and data-driven manner. AI-guided response prediction could estimate the likelihood of benefit before implantation. AI-driven network analyses could identify optimal stimulation targets based on circuit influence rather than seizure onset alone. AI-based state estimation could enable continuous forecasting of seizure risk across short-term, circadian, and multidien timescales, supporting stimulation during periods of rising vulnerability rather than solely in response to detected seizures. Also, AI-enabled adaptive control could optimize stimulation parameters through state-dependent strategies that balance efficacy, safety, tolerability, and energy constraints as brain dynamics evolve. This review synthesizes biological, clinical, and computational advances to outline a roadmap for safe, interpretable, and individualized AI-guided neuromodulation in DRE.