Evaluation of Deep Learning Models on Alzheimer's MRI Dataset: AD-VGG16, AD-Resnet50, and AD-2DCNN

Alzheimer's disease accounts for 60-70% of instances of dementia. It is a neurological illness that often begins slowly, progresses, and worsens over time. The most typical early symptom is trouble memorizing recent events. As the illness progresses, symptoms may include confusion, difficulty speaking, and difficulty doing daily tasks. Using MRI medical images, previous studies have considered the application of deep learning to Alzheimer's disease, but this study introduced a novel aspect by simultaneously evaluating AD-VGG16, AD-Resnet50, and AD-2DCNN. Synergistic evaluation of these models provides a unique perspective to improve diagnostic accuracy. We used various preprocessing techniques, including pixel normalization and data augmentation using a well-known Alzheimer's MRI dataset, such as the Alzheimer MRI Preprocessed Dataset from Kaggle, to evaluate the models' precision, area under the curve (AUC), accuracy, and recall. Our analysis reveals distinct performance levels among the models. AD-VGG16, AD-Resnet50, and the 2D CNN were evaluated with resulting accuracies of 93.45%, 79.43%, and 98.28% on the test set, respectively. This research discusses a comparative evaluation of three frequently used deep learning models for the classification of Alzheimer's disease.