Multimodal neuroimaging fusion in presurgical evaluation of drug-resistant epilepsy: current evidence, clinical integration, and future directions
TL;DR
It is argued that the central challenge of the field is not the proliferation of techniques but clinical translation: improved localization confidence does not automatically improve seizure freedom, and reported performance figures—whether for multimodal concordance or for artificial-intelligence-based lesion detection—derive largely from retrospective, single-center, or internally validated cohorts and are best interpreted as upper bounds rather than as accuracy deployable across centers.
Abstract
Approximately 30% of patients with epilepsy are drug-resistant, and for many of them surgical resection of the epileptogenic zone (EZ) offers the best chance of seizure freedom. Because the EZ cannot be measured directly by any single test, it must be inferred from the convergence of complementary imaging markers, and multimodal neuroimaging fusion has therefore become central to modern presurgical evaluation. This critical narrative review examines what each modality—structural MRI, [18F]FDG-PET, ictal SPECT/SISCOM, magnetoencephalography and electric source imaging, functional MRI, diffusion tensor imaging, and arterial spin labeling—contributes as a surrogate marker of the epileptic network, and how these are combined through co-registration, three-dimensional integration, hybrid PET/MRI, quantitative methods, and machine learning. Rather than cataloging modalities, we organize the clinical evidence by clinical scenario—including MRI-negative and extratemporal epilepsy, suspected focal cortical dysplasia, discordant non-invasive data, and stereo-EEG planning—and provide a practical framework matching modality combinations to specific clinical decisions. We argue that the central challenge of the field is not the proliferation of techniques but clinical translation: improved localization confidence does not automatically improve seizure freedom, and reported performance figures—whether for multimodal concordance or for artificial-intelligence-based lesion detection—derive largely from retrospective, single-center, or internally validated cohorts and are best interpreted as upper bounds rather than as accuracy deployable across centers. Network-based approaches and personalized computational models, including the Virtual Epileptic Patient, are conceptually powerful but remain largely investigational, and we conclude that multimodal imaging augments rather than replaces expert multidisciplinary evaluation. Prospective multicenter validation and workflow standardization remain the key priorities for clinical translation.