Automated analysis of FDG PET as a tool for single-subject probabilistic prediction and detection of Alzheimer's disease dementia.Eur J Nucl Med Mol Imaging. 2013 Sep;40(9):1394-405
Authors: Arbizu J, Prieto E, Martínez-Lage P, Martí-Climent JM, García-Granero M, Lamet I, Pastor P, Riverol M, Gómez-Isla MT, Peñuelas I, Richter JA, Weiner MW, Alzheimer's Disease Neuroimaging Initiative
PURPOSE: To introduce, evaluate and validate a voxel-based analysis method of ¹⁸F-FDG PET imaging for determining the probability of Alzheimer's disease (AD) in a particular individual.
METHODS: The subject groups for model derivation comprised 80 healthy subjects (HS), 36 patients with mild cognitive impairment (MCI) who converted to AD dementia within 18 months, 85 non-converter MCI patients who did not convert within 24 months, and 67 AD dementia patients with baseline FDG PET scan were recruited from the AD Neuroimaging Initiative (ADNI) database. Additionally, baseline FDG PET scans from 20 HS, 27 MCI and 21 AD dementia patients from our institutional cohort were included for model validation. The analysis technique was designed on the basis of the AD-related hypometabolic convergence index adapted for our laboratory-specific context (AD-PET index), and combined in a multivariable model with age and gender for AD dementia detection (AD score). A logistic regression analysis of different cortical PET indexes and clinical variables was applied to search for relevant predictive factors to include in the multivariable model for the prediction of MCI conversion to AD dementia (AD-Conv score). The resultant scores were stratified into sixtiles for probabilistic diagnosis.
RESULTS: The area under the receiver operating characteristic curve (AUC) for the AD score detecting AD dementia in the ADNI database was 0.879, and the observed probability of AD dementia in the six defined groups ranged from 8% to 100% in a monotonic trend. For predicting MCI conversion to AD dementia, only the posterior cingulate index, Mini-Mental State Examination (MMSE) score and apolipoprotein E4 genotype (ApoE4) exhibited significant independent effects in the univariable and multivariable models. When only the latter two clinical variables were included in the model, the AUC was 0.742 (95% CI 0.646 - 0.838), but this increased to 0.804 (95% CI 0.714 - 0.894, bootstrap p=0.027) with the addition of the posterior cingulate index (AD-Conv score). Baseline clinical diagnosis of MCI showed 29.7% of converters after 18 months. The observed probability of conversion in relation to baseline AD-Conv score was 75% in the high probability group (sixtile 6), 34% in the medium probability group (merged sixtiles 4 and 5), 20% in the low probability group (sixtile 3) and 7.5% in the very low probability group (merged sixtiles 1 and 2). In the validation population, the AD score reached an AUC of 0.948 (95% CI 0.625 - 0.969) and the AD-Conv score reached 0.968 (95% CI 0.908 - 1.000), with AD patients and MCI converters included in the highest probability categories.
CONCLUSION: Posterior cingulate hypometabolism, when combined in a multivariable model with age and gender as well as MMSE score and ApoE4 data, improved the determination of the likelihood of patients with MCI converting to AD dementia compared with clinical variables alone. The probabilistic model described here provides a new tool that may aid in the clinical diagnosis of AD and MCI conversion.