Clinical validation of a Grid-based SPM web tool for the automatic assessment of [18F]FDG-PET brain metabolic automatic assessment of [18F]FDG-PET brain metabolic

Background: [18F]FDG-PET imaging has been recently suggested to increase diagnostic accuracy in neurodegenerative disorders since the very early stages (Dubois et al., 2010; Jack et al., 2011; Albert et al., 2011; Sperling et al., 2011; Ravskowski et al., 2011). At present, however, glucose metabolism for diagnostic purposes is mostly evaluated through visual inspection of [18F]FDG-PET brain images in single subjects, thus resulting in low sensitivity and specificity. Aim: A clinical validation of objective voxel-based maps of hypometabolism generated through statistical parametric mapping (SPM) using [18F]FDG-PET scans in a large series of patients with neurodegenerative disorders (e.g. Mild Cognitive Impairment-MCI, probable Alzheimer's disease-AD and Frontotemporal Lobar Degeneration-FTLD) by comparing single-subject [18F]FDG-PET scans to a large population of normal subjects. We aimed at providing a tool with high statistical power and high sensitivity and specificity for early and differential diagnosis of neurodegenerative disorders. Materials and Methods: 112 normal scans controls were included in a database for single-patient analysis. All included images underwent quality control procedures, including two-pass masked-normalization, smoothing, intensity rescaling, global count intensity normalization and distance analysis. The influence of scanner effects and demographic variables was measured as well. Glucose metabolism was then investigated in 95 patients with a clinical diagnosis of neurodegenerative disease. Visual ratings of resulting SPM maps of glucose hypometabolism were provided by a team of experienced neurologists. Results: The comparison of a single case against a large group of controls yielded SPM Maps of hypometabolism with high t-values corrected for multiple comparisons (FWE) at the voxel level. Analysis of raters' performance for diagnostic accuracy of FDG-PET provided very high sensitivity values (higher than 95%) significantly increasing the level of diagnostic confidence with respect to the FDG visual inspection. In particular, AD and FTLD patterns were differentiated and in the case of amnestic MCI, the results were either negative or showed and AD pattern. Conclusions: This voxel-based FDG-PET tool can increase both statistical significance and diagnostic confidence when evaluating specific hypometabolic patterns in single subjects for earlier and more accurate disease detection. Furthermore, this tool may be easily implemented as a Grid-web service (Castiglioni et al., 2009) representing an extremely powerful toll for clinicians on standard workstations in order to obtain a disease confirmatory or exclusionary tests.