Alzheimer's disease is a progressive neurologic disorder that causes the brain to shrink (atrophy) and brain cells to die. A recent study found that 40 million people worldwide suffer from Alzheimer's disease (AD). A few symptoms of this AD disease are problems with language understanding, mood swings, behavioral issues, and short-term memory loss. A key research area for AD is the classification of stages. In this paper, we applied both binary and multi-class classification. In this paper, proposed is a Mask-Region based Convolution Neural Network (R-CNN) for classifying the stages including MCI, LMCI, EMCI, AD, and CN of Alzheimer's Disease. First performing pre-processing by using the skull-stripping algorithm for removing the noise. Second, the patch wise U-Net has been employed to segment the images for improving the classification process. After that, the system's efficiency is examined using MATLAB-based experiments, utilizing images from the Alzheimer's disease Neuroimaging Initiative (ADNI) dataset to evaluate the efficiency in terms of accuracy, precision, recall, specificity, and sensitivity. Our proposed approach to classifying the stages achieves about 98.54%,94.2%, 98.25%, 99.2%, and 99.02%in terms of accuracy with EMCI, CN, MCI, AD, and LMCI respectively. Proposing mask R-CNN with segmentation to classify from CN to AD subjects successfully improved classifier accuracy significantly on the ADNI datasets.