Comparison of Functional Connectivity-Based Graph Metrics in Pre-Surgical Planning for Drug-Resistant Epilepsy

Abstract

Patients with drug-resistant epilepsy often require surgical treatment to reduce seizure frequency. The challenges posed by the network nature of epilepsy and the difficulty of objectively locating the epileptogenic zone (EZ) create a need for improved tools to characterize pathological network dynamics. Graph-theoretical metrics derived from functional connectivity (FC) analysis may support surgical planning by enabling quantitative assessment of regional epileptogenicity in the brain and improving postoperative outcome prediction.

This thesis compares findings across relevant studies to evaluate the consistency and robustness of FC-derived graph metrics in identifying the EZ and predicting postoperative outcomes. The performance of the graph metrics degree centrality (DC), nodal strength (NS), betweenness centrality (BC), and eigenvector centrality (EVC) is examined in relation to epileptogenic zone discrimination and postoperative outcome prediction. Metric performance is quantified with measures such as area under the curve (AUC) and positive predictive value (PPV). The clinical implications of these results, as well as related methodological limitations, are also discussed.

Overall, the reviewed metrics demonstrate meaningful value for EZ discrimination and surgical outcome prediction, particularly when optimal connectivity methods and frequency bands are used. Metrics derived from amplitude envelope correlation (AEC) in the gamma band showed consistently strong performance across the reviewed studies. The highest performance values were reported for NS derived from AEC (EZ discrimination AUC = 0.81 and outcome prediction PPVs 75–82 %) and Pearson correlation (Outcome prediction AUC ≈ 0.9), as well as for AEC-based EVC (EZ discrimination and outcome prediction PPVs 78–83 %). Notably, all metrics demonstrated moderate to strong performance in EZ discrimination and outcome prediction, even when only interictal data was analyzed.

The reviewed results indicate that FC-based graph metrics hold promising value for informing surgical planning, but their performance is sensitive to methodological choices. The heterogenous patient cohorts, analysis conditions and metric performance evaluation criteria limit direct comparison between the results obtained across reviewed studies. With further standardization and result validation in larger patient cohorts, the analysis of FC-based graph metric values may become an increasingly significant tool in supporting surgical planning for patients with drug-resistant epilepsy.