Alzheimer’s disease (AD) is a chronic neurodegenerative disease of the central nervous system that has no cure
and leads to death. One of the most prevalent tools for AD diagnosis is magnetic resonance imaging (MRI), because of its
capability to visualize brain anatomical structures. There is a variety of classification methods for automatic diagnosis of
AD, such as support vector machines, genetic algorithms, Bayes classifiers, neural networks, random forests, etc., but
none of them provides robust information about the stage of the AD, they can just reveal the presence of disease.
In this paper, a new approach for classification of MRI images using a fuzzy inference system is proposed. Two statistical
moments (mean and standard deviation) of 116 anatomical regions of interests (ROIs) are used as input features for the
classification system. A t-test feature selection method is used to identify the most discriminative ROIs. In order to evaluate
the proposed system, MRI images from a database consisting of 818 subjects (229 normal, 401 mild cognitive impairment
and 188 AD subjects) collected from the Alzheimer’s disease neuroimaging initiative (ADNI) is analyzed. The
receiver operating characteristics (ROC) curve and the area under the curve (AUC) of the proposed fuzzy inference system
fed by statistical input features are employed as the evaluation criteria with k-fold cross validation. The proposed system
yields promising results in normal vs. AD classification with AUC of 0.99 on the training set and 0.8622±0.0033 on
the testing set.