Combining MRI and CSF measures for classification of Alzheimer's disease and prediction of mild cognitive impairment conversion

The suggested revision of the NINCDS-ADRDA criterion for the diagnosis of Alzheimer's disease (AD) includes at least one abnormal biomarker among magnetic resonance imaging (MRI), positron emission tomography (PET) and cerebrospinal fluid (CSF). We aimed to investigate if the combination of baseline MRI and CSF could enhance the classification of AD compared to using either alone and predict mild cognitive impairment (MCI) conversion at multiple future time points. 369 subjects from the Alzheimer's disease Neuroimaging Initiative (ADNI) were included in the study (AD=96, MCI=162 and CTL=111). Freesurfer was used to generate regional subcortical volumes and cortical thickness measures. A total of 60 variables were used for orthogonal partial least squares to latent structures (OPLS) multivariate analysis (57 MRI measures and 3 CSF measures: Aβ(42), t-tau and p-tau). Combining MRI and CSF gave the best results for distinguishing AD vs. CTL. We found an accuracy of 91.8% for the combined model at baseline compared to 81.6% for CSF measures and 87.0% for MRI measures alone. The combined model also gave the best accuracy when distinguishing between MCI vs. CTL (77.6%) at baseline. MCI subjects who converted to AD by 12 and 18month follow-up were accurately predicted at baseline using an AD vs. CTL model (82.9% and 86.4% respectively), with lower prediction accuracies for those MCI subjects converting by 24 and 36month follow up (75.4% and 68.0% respectively). The overall prediction accuracies for converters and non-converters ranged from 58.6% to 66.4% at different time points. Combining MRI and CSF measures in a multivariate model at baseline gave better accuracy for discriminating between AD and CTL, between MCI and CTL and for predicting future conversion from MCI to AD, than using either MRI or CSF separately.