[11C]Flumazenil PET in temporal lobe epilepsy: do we need an arterial input function or kinetic modeling?

Reduced signal on [(11)C]]flumazenil (FMZ) positron emission tomography (PET) is associated with epileptogenic foci. Linear correlations within individuals between parametric and nonparametric images of FMZ binding have been shown, and various methods have been used, without comparison of diagnostic usefulness. Using hippocampal sclerosis (HS) as a test case, we formally compare the diagnostic yield of parametric images obtained either with a parent tracer arterial plasma input function and spectral analysis (yielding volume-of-distribution (VD) images), or with an image-based input function and the simplified reference tissue model (binding potential images, BP-SRTM) with the diagnostic yield of semiquantitative-integrated (ADD) images from 10 to 20 or 20 to 40 mins (ADD1020 and ADD2040). Dynamic 90-min [(11)C]FMZ PET datasets and arterial plasma input functions were available for 15 patients with medically refractory medial temporal lobe epilepsy (TLE) and histologically verified unilateral HS and for 13 control subjects. SPM2 was used for analysis. ADD1020 and ADD2040 images showed decreased FMZ uptake ipsilateral to the epileptogenic hippocampus in 13/15 cases; 6/13 had bilateral decreases in the ADD1020 analysis and 5/13 in the ADD2040 analysis. BP-SRTM images detected ipsilateral decreases in 12/15 cases, with bilateral decreases in three. In contrast, VD images showed ipsilateral hippocampal decreases in all 15 patients, with bilateral decreases in three patients. Bilateral decreases in the ADD images tended to be more symmetrical and in one case were more marked contralaterally. Full quantification with an image-independent input should ideally be used in the evaluation of FMZ PET; at least in TLE, intrasubject correlations do not predict equivalent clinical usefulness.