Introduction: Statistical parametric mapping (SPM) provides useful voxel-by-voxel analyses of brain images from (18)F-fluorodesoxyglucose positron emission tomography (FDG-PET) after an initial step of spatial normalization through an anatomical template model. In the setting of the preoperative workup of patients with temporal epilepsy, this study aimed at assessing a block-matching (BM) normalization method, where most transformations are computed through small blocks, a principle that minimizes artefacts and overcomes additional image-filtering.
Methods: Brain FDG-PET images from 31 patients with well-characterised temporal lobe epilepsy and among whom 22 had common mesial temporal lobe epilepsy were retrospectively analysed using both BM and conventional SPM normalization methods and with PET images from age-adjusted controls. Different threshold p values corrected for cluster volume were considered (0.01, 0.005, and 0.001).
Results: The use of BM provided equivalent values to those of SPM with regard to the overall volumes of temporal and extra-temporal hypometabolism, as well as similar sensitivity for detecting the involved temporal lobe, reaching 87 and 94 % for SPM and BM, respectively, at a threshold p value of 0.01. However, the ability to more accurately localize brain lesions within the mesial portion of the temporal lobe was a little higher with BM than with SPM with respective sensitivities reaching 78 % for BM and 45 % for SPM (p < 0.05).
Conclusions: BM normalization compares well with conventional SPM for the voxel-based quantitative analysis of the FDG-PET images from temporal epilepsy patients. Further studies in different population are needed to determine whether BM is truly an accurate alternative to SPM in this setting.
Keywords: Block-matching algorithm; FDG-PET; Mesial temporal lobe epilepsy; Spatial normalization; Statistical parametric mapping.