Alzheimer's disease affects irreversible brain damage cells that control memory and reasoning. Owing to comparable brain patterns and pixel intensities, diagnosing Alzheimer's disease in elderly persons is challenging and necessitates a highly discriminative feature representation for classification. These representations can be learned from data using deep learning algorithms. To address these issues, previous research looks into the usefulness of rs-fMRI for multi-class classification of Alzheimer's disease and its phases, such as CN, SMC, EMCI, MCI, LMCI, and AD. To improve the image quality, some pre-processing procedures were used, including brain extraction, motion correction, intensity, normalization, spatial smoothing, high pass filtering, spatial normalizing, image registration, and 4D to 2D conversion. And then Alzheimer’s disease multi stage classification was done by using Alex net and Resnet convolutional neural network. Existing research, on the other hand, hasn't concentrated on texture feature extraction. ResNet has proven to be successful in a wide range of applications, but one major drawback is that training a deeper network takes several days, making it unsuitable for real-world applications. Medical records, whether housed in health information systems, the cloud, etc. are critical. For such records, privacy and security must be ensured using encryption and authentication techniques that are not currently in use. To prevent these issues, this study used pre-processing techniques as brain extraction, motion correction, intensity, normalization, spatial smoothing, high pass filtering, spatial normalizing, image registration, and 4D to 2D conversion to improve image quality. And then texture feature extraction will be done by utilizing gray level co occurrence matrix (GLCM). Alzheimer's disease multi stage classification will be computed based on Alexnet and Googlenet models for Neuroimaging data is attained from Alzheimer’s disease Neuroimaging Initiative (ADNI) dataset. And then all Medical records will be encrypted using Modified Advanced Encryption Standard (MAES) to provide security and then stored in cloud. Experiments show that the suggested model is effective with respect to encryption time, precision, recall, and accuracy.