Accuracy of spatial normalization of the hippocampus: implications for fMRI research in memory disorders

Functional magnetic resonance imaging (fMRI) studies in memory impairment have detected functional alterations in medial temporal lobe (MTL) structures, notably the hippocampus. Many of these studies employ spatial normalization to place subjects in a standardized template space prior to analysis; however, little is known about the effects of local atrophy on the normalization process in structures such as the hippocampus. The purpose of this study was to compare the accuracy of spatial normalization of the hippocampus between memory-impaired patients and controls. Twenty clinically-defined mild cognitive impairment (MCI) subjects and twenty elderly controls were studied at 4T with structural and functional MRI during a memory encoding-retrieval task. Bilateral hippocampal regions-of-interest (ROIs) were manually drawn for all subjects and further divided into anterior/posterior subregions. To assess normalization accuracy to the Montreal Neurological Institute template, the percentage of each template-defined hippocampal ROI originating from true hippocampal tissue was determined for all subjects. To assess the ability of spatial normalization to equalize group differences in hippocampal volume, pre- and post-normalization hippocampal volumes were compared. Finally, fMRI measures from template and non-template analyses were compared. Poorer normalization accuracy of the bilateral hippocampi, particularly the posterior portions, was found for MCI subjects. Significant group differences were found in left hippocampal and bilateral posterior hippocampal volumes, and these differences were not corrected with normalization. Hippocampal volumes were significantly correlated with normalization accuracy across MCI and control groups, but some significant differences in normalization accuracy persisted independent of these volume differences. Template and non-template fMRI analyses were significantly correlated in controls, but not MCI subjects, during memory retrieval. These findings suggest decreased normalization accuracy in memory-impaired subjects is a potentially important confounder of template-based fMRI analyses in the hippocampus and MTL.