A Deployment-Oriented Review of EEG Signal Processing, Network Analysis, and Neuromodulation in Epilepsy

Abstract

Approximately one third of patients with epilepsy remain refractory to pharmacological treatment. This creates a strong clinical need for methods that can support diagnosis, monitoring, and intervention under real clinical conditions. Despite substantial recent progress in epilepsy-related technologies and algorithms, strong retrospective performance often fails to translate into reliable chronic deployment because of drift, class imbalance, stimulation artifacts, and limited or absent ground-truth feedback.

This review provides a deployment-oriented synthesis of epilepsy-related methods across the sensing-to-control pipeline, spanning data acquisition, preprocessing, spontaneous event and evoked-response analysis, connectivity analysis, brain modeling, and closed-loop control. It is based on a structured scoping search, with studies selected and synthesized according to deployment relevance, validation setting, and role within the sensing-to-control pipeline. The methods are organized using a shared four-axis framework: temporal scale and latency constraints, observability and representation, vulnerability to non-stationarity and drift, and deployment role. This framework highlights how upstream constraints shape the reliability of downstream inference and intervention.

We argue that algorithm-guided neuromodulation in epilepsy is best understood as a constrained control problem under partial observability and non-stationarity. Within this framing, we revisit three recurring questions central to system design: whether seizure prediction is achievable, the validation status of candidate biomarkers, and the balance between focal and network models of epileptogenicity.

For each pipeline stage, we identify reporting and evaluation priorities aligned with clinically relevant endpoints rather than offline accuracy alone. Together, this review highlights engineering priorities for clinically durable systems and generates testable hypotheses about seizure dynamics and network controllability.