Relating one-year cognitive change in mild cognitive impairment to baseline MRI features

Background: We propose a completely automated methodology to investigate the relationship between magnetic resonance image (MRI) features and changes in cognitive estimates, applied to the study of MiniMental State Examination (MMSE) changes in mild cognitive impairment (MCI). Subjects: A reference group composed of 75 patients with clinically probable Alzheimer's Disease (AD) and 75 age-matched controls; and a study group composed of 49 MCI, 20 having progressed to clinically probable AD and 29 having remained stable after a 48 month follow-up. Methods: We created a pathology-specific reference space using principal component analysis of MRI-based features (intensity, local volume changes) within the medial temporal lobe of T1-weighted baseline images for the reference group. We projected similar data from the study group and identified a restricted set of image features highly correlated with one-year change in MMSE, using a bootstrap sampling estimation. We used robust linear regression models to predict one-year MMSE changes from baseline MRI, baseline MMSE, age, gender, and years of education. Results: All experiments were performed using a leave-one-out paradigm. We found multiple image-based features highly correlated with one-year MMSE changes (|r|N0.425). The model for all N= 49 MCI subjects had a correlation of r= 0.31 between actual and predicted one-year MMSE change values. A second model only for MCI subjects with MMSE loss larger than 1 U had a pairwise correlation r= 0.80 with an adjusted coefficient of determination r 2= 0.61. Findings: Our automated MRI-based technique revealed a strong relationship between baseline MRI features and one-year cognitive changes in a sub-group of MCI subjects. This technique should be generalized to other aspects of cognitive evaluation and to a wider scope of dementias