Abstarct: Alzheimer’s disease (AD) is the most common neurodegenerative disorder that impairs memory, cognitive function, behavior and language. There is no cure for AD yet but, early detection of AD prevent the symptoms of AD and slow down its progression. Recently, the positron emission tomography (PET), that measures, in particular, the brain’s rate of glucose mextabolism and also is more accurete than structural neuroimaging modalities, e.g. magnetic resonance imaging (MRI), increasingly is used to investigate the prognosis and diagnosis of AD. In this thesis, an automated approach for AD diagnosis baxsed on local intensity voxel histograms (IVH) and gray level co-occurrence matrices (GLCM) features are proposed. Local IVH and GLCM can provide statistical and texture information of 48 regions of interest from PET neuroimaging. Moreover, a reliable feature baxsed on local IVHs is proposed. We found that classification accuracies baxsed on different feature analysis of the ROIs varied from 72.9% to 88%. Using the proposed feature yields a classification accuracy of 86.3% (sensitivity = 80.8% and specificity = 90.5%) and outperformed local IVH baxsed features that its highest accuracy is 84.5%. In addition, we propose a combination of outputs of top five features to discriminate between AD and normal controls, using fuzzy-rule-baxsed classifier. The proposed method uses the shuffled frog leaping algorithm (SFLA) and genetic algorithm operators to extract a set of fuzzy membership functions and the size and structure of fuzzy rules, named GA-SFLA. The characteristics of GA-SFLM is that optimal parameters of the fuzzy classifier are extracted from the training data using SFLA. The obtained classification accuracy for AD diagnosis is 93.33%, which reveals that proposed method outperforms several recent methods.