Integration of cognitive tests and resting state fMRI for the individual identification of Mild Cognitive Impairment

Background: Resting state functional magnetic resonance imaging (RS-fMRI) is a non-invasive and in vivo technique consisting in the acquisition of blood oxygen level-dependent (BOLD) data in the absence of stimulation by a task, which allows the study of the brain functional networks independently of any task. It has been suggested as a promising imaging technique to find early biomarkers of neurodegenerative disorders, which can be more sensitive to earlier stages of disease than structural alterations. Recent findings have highlighted the potential usefulness of this technique for the early diagnosis of amnestic mild cognitive impairment (aMCI), the prodromal stage of Alzheimer's disease. It is not yet established, however, whether RS-fMRI adds any quantitative predictive/classificatory value to that achieved with standard cognitive tests at the individual level. Methods: We present a systematic analysis of the impact of different RS-fMRI derived indices in the classification procedure. A selection of 400 different features were extracted from RS-fMRI of 29 aMCI patients and 21 age-matched control subjects. We also had scores from 21 cognitive tests available for each patient/subject. Using standard machine learning algorithms we computed the most relevant features for aMCI classification considering RS-fMRI data or cognitive test scores alone, and then these were combined. Finally, we evaluated the classification performance for these features using a Monte Carlo 10-fold cross validation analysis. Results: We obtained an accuracy (sensitivity/specificity/area under curve) of 0.9319 (0.9258/0.9415/0.9454) when using the cognitive test scores only, 0.9541 (0.9582/0.9485/0.9918) when considering RS-fMRI features only, and 0.9559 (0.9620/0.9470/0.9517) when using both sets of measure. Scores on Category Fluency and Rey delayed Memory where among the most useful cognitive classifiers, whilst indices of local connectivity (in a neighbouring area of 10mm) and of connectivity of the left posterior superior temporal gyrus, the right anterior superior temporal gyrus, and the left superior frontal gyrus where the most useful classifiers among the RS-fMRI indices. Conclusions: Our results demonstrate that RS-fMRI provides complementary information to cognitive tests for aMCI-patients/subjects classification, mostly related to local connectivity information and the correlation of functional activity of the superior temporal gyrus