Alzheimers Disease Recognition using CNN Model with EfficientNetV2

Abstract

In recent years, machine learning-based identification of Alzheimer's disease (AD) from the use of brain images such as MRI has been a hot topic. Deep learning's recent triumph in object recognition has stimulated related research. Such algorithms, however, have some disadvantages, including the requirement for a large number of training pictures and rigorous deep network design optimization. To address these issues, we employ CNN and ANN in this study. Weights are pre-trained on huge natural picture benchmark datasets in state-of-the-art designs like EfficientNetV2, and only a few numbers of MRI images are used to retrain the comprehensive layers in state-of-the-art architectures like EfficientNetV2. Our dataset contains 4 characteristics, as well as 5121 training shots and 1279 testing photos. The data was initially divided into three sets: training, validation, and test, with just the training, validation sets being used to select models. To minimize overfitting, the tests were abandoned undisturbed until the collaborative process was completed. The various techniques performed well when applied to when using datasets with alternative criteria for inclusion or demographic features, this is not the case.