OBJECTIVES

Localization of the network underlying drug-resistant focal epilepsy in individuals considering surgical treatment with unremarkable MRI is challenging. Concordance rates of 40%-69% have been reported with FDG-PET image statistical parametric mapping (SPM). We investigated the efficacy of postprocessing specific to cortices by cortex-based mapping (CBM) on hybrid PET/MR images with healthy subjects to localize sites of seizure onset.

METHODS

We retrospectively examined the PET/MR images of 42 MRI-negative individuals with drug-resistant focal epilepsy who had surgery and 23 healthy subjects. Visual interpretation of standardized uptake value ratios (SUVRs), voxelwise mapping with a two-sample t-test of SUVRs (t-map, SPM), and the proposed z-transformation of the SUVR of patients compared with those of healthy subjects acquired with CBM were compared with the surgical field. Kappa tests, conclusive concordance (CC), partial concordance (PC), and discordance were estimated, with McNemar's test determining the superiority.

RESULTS

After an average follow-up of 37.2 months, in people who were seizure-free (n = 31; functionally silent cortices in 26), the CC rate with CBM was 87.10%. Performance was CBM (CC:PC = 27:1), t-map (CC:PC = 15:1), and SUVR (CC:PC = 0:17). The sensitivity, specificity, and kappa scores were 0.87, 0.91, and 0.717 (p < 0.001) for CBM and 0.48, 0.73, and 0.153 (p = 0.288) for t-maps, respectively. The CBM approach was superior to the t-map (p < 0.001) in most extratemporal epilepsies. The average Pearson's r of CBM and t-map to artifacts was 0.08 ± 0.02 and 0.33 ± 0.02, respectively.

INTERPRETATION

By eliminating intersubject morphological variations and explicit statistics at the cortex, CBM localized the seizure origin in MRI-negative epilepsy patients with superior efficiency.