Seizure-onset zone localization by statistical parametric mapping in visually normal (18) F-FDG PET studies

Objective: Neuroimaging is crucial in the presurgical evaluation of patients with medically refractory epilepsy. To improve the moderate sensitivity of [(18) F]fluorodeoxyglucose-positron emission tomography ((18) F-FDG-PET), our aim was to evaluate the usefulness of statistical parametric mapping (SPM) to localize the seizure-onset zone (SOZ) in PET studies deemed normal by visual assessment.

Methods: Fifty-five patients with medically refractory epilepsy whose (18) F-FDG-PET was visually evaluated as normal were retrospectively included. Twenty of these patients had undergone surgical intervention. PET images were analyzed by SPM8 using a corrected p-value of p < 0.05 and three uncorrected p-values of p < 0.0001, p < 0.001, and p < 0.005, matched with minimum cluster sizes of k > 0, k > 20, k > 100, and k > 200, respectively. The SPM-identified potential seizure zone (SZ) was compared to the SOZ, which was determined by consensus during patient management meetings in the epilepsy unit, taking into account presurgical tests. Studies in which the SPM-identified potential SZ was concordant with the SOZ were considered "correctly localizing."

Results: The SPM threshold combination with the least restrictive p-value and greatest minimum cluster size achieved the highest rate of correctly localizing studies. When p < 0.005/k > 200 was used, 40% (22/55) of studies were correctly localizing, and the concordance obtained in the surgically intervened subgroup was substantial (к = 0.607, 95% confidence interval [CI] 0.258-0.957), which was comparable to the concordance obtained by magnetic resonance imaging (MRI) (к = 0.783, 95% CI 0.509-1.000).

Significance: SPM offers improved SOZ localization in (18) F-FDG-PET studies that are negative on visual assessment. For this purpose, statistical parametric maps could be thresholded with liberal p-values and restrictive cluster sizes.