In this study, we modeled each anatomical region in the human brain as a Shannon information source using functional Magnetic Resonance Images (fMRI). First, we estimated the probability density functions of the regions by considering the voxel time series in the anatomical regions as random variables. Then, using these probability density functions, we estimated the entropy of the regions and the Kullback-Leibler (KL) divergence between regions. Based on the suggested model, we created two types of feature spaces. We defined entropy vectors by adding the entropy values we calculated for each anatomical region in the first feature space under a vector. In the second feature space, we define KL vectors, whose elements are KL divergences. In order to show the validity of the suggested brain model, we test the performance of multilayer perceptron on the ADNI (Alzheimer's Disease Neuroimaging Initiative) dataset. Multilayer Perceptrons are trained with entropy and KL vectors obtained from the fMRI images of these subjects and promising results were obtained for early diagnosis of the disease.