MRI‐negative epilepsy: A systematic review and meta‐analysis

Drug-resistant focal epilepsy is commonly dichotomized based on magnetic resonance imaging (MRI) lesion visibility into positive (MRI-pos) and negative (MRI-neg). Yet, the criteria used to ascribe such categorization are variable. We used a systematic review and meta-analysis to synthesize evidence for the designation of MRI-neg status. In accordance with Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines, the systematic review (1990-2025) across Embase, Cochrane, and Medline databases identified cohorts with MRI-neg epilepsy. Unsupervised clustering stratified studies based on co-occurrence of imaging modalities. Within identified classes, we assessed the consistency of reporting MRI parameters, rater expertise, post-processing, and stereo-electroencephalography (SEEG). Meta-analyses evaluated the effects of post-processing on diagnostic yield and MRI-neg status on post-surgical outcome. We screened 2622 records and assessed the eligibility of 448 full-text articles, 246 of which met the inclusion criteria for systematic review: 108 (44%) provided data only on MRI-neg and 138 (56%) on mixed adult cohorts, for a total of 10.463 MRI-neg and 7436 MRI-pos patients. Compared to MRI-pos, MRI-neg patients underwent SEEG more frequently (75% vs 54%, p < 0.05), underwent surgery less frequently (73% vs 84%; odds ratio [OR] = 1.14, p < 0.001), and had less favorable outcomes (61% vs 72%, p < 0.05). Clustering identified three classes: MRI-dominant, typified by consistent reporting of MRI parameters (ORs >3.11, p < 0.001), rater-expertise (ORs >9.94, p < 0.001), and post-processing (ORs >3.38, p < 0.03), as opposed to Limited-MRI (χ2 = 41.08, p < 0.001); MRI-and-nuclear-imaging class was typified by use of SEEG (ORs >3.33, p < 0.02). Meta-analyses showed a 39% gain in diagnostic yield after post-processing (11.10, 95% confidence interval [CI] 7.45-16.53) and a higher proportion of favorable surgical outcome in MRI-pos compared to MRI-neg (75% vs 58%; χ2 = 19.10, p < 0.001). Time-based sensitivity analyses did not affect results. The designation of MRI-neg is ambiguous, with most studies lacking details on imaging parameters and reader expertise. Given a 39% gain in diagnostic yield, MRI post-processing should be performed systematically as part of a modern multimodal approach to epilepsy surgery before ascribing MRI-neg status.