A ROC analysis of hippocampal volumetric MRI (VMRI) in determining Alzheimer's disease (AD) in patients with positive FDG brain PET (PET)

Objectives To compare PET with VMRI in the evaluation of demented patients.

Methods 59 subjects with clinical diagnosis of dementia were studied with PET brain scans using visual and 3DSSP quantitation. MRI including volumetric analysis was also performed. A subset of 26 patients with a high degree of suspicion of AD based on clinical assessment and PET findings of biparietal, posterior temporal, and posterior cingulate relative metabolic reductions of greater than 2SD were considered as high probability for AD (AD “Gold Standard”). All others were classified as Non AD. The hippocampal volume percentile distributions of the AD and Non AD groups were compared. The sensitivity and specificity of VMRI was calculated at multiple hippocampal percentile cutoffs and a ROC analysis was performed.

Results Using a 2SD cutoff, VMRI demonstrated volume reduction in 7 of 28 “Gold Standard” AD patients (Sens = 25%) and VMRI volume was above the cutoff in 27 of 31 Non AD patients (Spec = 87%). The ROC analysis determined an Area Under the Curve = 0.52. Of the Non AD group, PET interpretation was normal in 3%, while findings consistent with atrophy (confirmed on MRI) were found in 84% and small vessel ischemic disease in 58%, also confirmed on MRI. 50% showed some other neuropathologic process as referenced in Table 1.

Conclusions In this series, the ROC analysis suggests that VMRI correlates poorly with the presence of AD. Hippocampal volumes should be looked upon with caution in attempting to diagnose AD in patients with dementia. FDG PET was able to characterize many of the Non AD patients as having other neuropathologic processes that may explain their dementia. A limitation of the study was the use of characteristic PET AD patterns as described in the literature along with clinical status as the gold standard for AD, however, tissue confirmation was not obtained.