Alzheimer's disease (AD) is an irreversible, complex mental disorder that affects individual thinking, memory, and skills. Many different statistical methods and algorithms are often employed to study neuroimaging data to classify Alzheimer patients and normal control subjects (NC). The resting-state functional resonance imaging (rs-fMRI) is an effective way to study AD and NC subjects. However, it is challenging to classify the different level of Alzheimer and normal control subjects based on rs-fMRI images because the brain activation is almost similar. The aim of this study is to classify different levels of Alzheimer patients and Normal control subjects by using a deep neural network (DNN) with TensorFlow and structural equation modeling (SEM). By implementing SEM structural relationships between brain regions are explored, and the impact of these relationships on predicting Alzheimer disease in patients was estimated. The result of SEM showed that AD mostly affects the Amygdala and Hippocampus regions than other parts of the brain. After exploring SEM model, we implement DNN with considering two categories of AD first and then considering multiple categories of AD. The test accuracy of trained DNN models in the aforementioned scenarios was achieved 96% and 98.6%, respectively. This proves the superiority and effectiveness of our proposed hybrid DNN model compared to the state-of-the-art.