Integrated Classification of Amnestic Mild Cognitive Impairment Using Functional and Structural MRI

As a prodromal phase of Alzheimer Disease (AD), amnestic Mild Cognitive Impairment (aMCI) may be the appropriate stage for clinical trials of early therapeutic intervention delaying AD progress. Blood Oxygen Level Dependent (BOLD) fMRI, as a non-invasive functional neuroimaging technique, has the potential to reflect the early neural network change associated with AD pathology at aMCI stage, and can be used to identify aMCI patients so as to enrich aMCI population for clinical trials. In the present study, we investigated how fMRI data could be used to differentiate potential aMCI patients from age-matched healthy controls. We explored the predictive power of multiple high dimensional fMRI measures acquired while subjects performed episodic memory encoding and recognition tasks, through the use of Support Vector Machine and Logistic Regression classifier. We compare the fMRI measure’s predicting power to the surface-based cortical structural measures and also investigated if integrating different fMRI measures and sMRI measures could improve the classifier performance. Our result demonstrate functional connectivity maps of hippocampus and inferior parietal cortex during encoding tasks achieved highest discriminative power (0.88), which is comparable to the accuracy achieved by the surface-based structural measure. Integrating two measures across different modalities (fMRI and sMRI) or from the same modality greatly increased the classification accuracy from 0.88 to over 0.96 (Leave-one-out-cross-validation). These results indicate combining high dimensional fMRI measures with dimensionality control method (Such as Principal Component Analysis) and machine-learning methods (such as Logistic Regression and Support Vector Machine) can possibly differentiate aMCI patients from control subjects with a high degree of accuracy. Furthermore, the functional and structural brain change of aMCI subjects reflected by different types of fMRI measures (i.e. functional contrasts and functional connectivity) and sMRI measures may be induced by asymmetrical Alzheimer’s pathological process. Integrating multiple measures can provide complementary information to classifiers and greatly increase classification accuracy. Furthermore, the most homogenous neural network feature for patients at aMCI stage may be the disconnection between hippocampus and prefrontal cortex. This hippo-frontal disconnection causes aMCI subjects’ difficulty in the formation of new memory and triggered extensive and more individual-specific compensational brain activation.