Optimized deep neural framework for multi-stage classification of Alzheimer's disease severity

Alzheimer's disease, a devastating neurodegenerative disorder, that necessitates a prompt and precise diagnosis to be properly handled. The present work introduces a deep learning technique for the identification of the ultimate patients’ Alzheimer stages by using MRI data. The proposed model utilizes the modern deep learning convolutional networks (CNN) architectures, such as InceptionV3, ResNet50, ResNet152 and Xception, for the multiclass classification task. To help the model to reach its best potential performance, a technique named ‘early stopping’ is used. Moreover, an ensemble model integrating CNN and NasNetMobile attains an accuracy of 0.992, being very efficient for the classification task. The YOLOv8 model represents a very strong tool for the detection of the entries of Alzheimer's disease from the MRI images, with a screening rate of 0.926. The results imply the advantages of the convolutional model-based methods specifically the convolutional neural networks ones, in the facilitation of the automated evaluation of Alzheimer's disease which is a valuable resource in medical decision-making and early therapeutic planning.