Recognition of Alzheimer’s Disease stages via InceptionV3 and ResNet50

Abstract

Early and precise detection of Alzheimer’s disease (AD) is essential for successful treatment. This research presents a system that autonomously detects and categorizes the phases of Alzheimer’s disease via brain scans and sophisticated deep learning techniques, including InceptionV3 and ResNet50. These models started with pretrained weights and were augmented by including bespoke classification layers, which consisted of dropout, batch normalization, and dense layers to increase performance and mitigate overfitting. The preprocessing processes included scaling the picture to 224 by 224 pixels, using average filtering for denoising, and converting the color space to guarantee compatibility with the models. Evaluations of the OASIS dataset illustrate the efficacy of the proposed approaches in accurately differentiating among the various phases of Alzheimer’s disease, including four classifications: nondemented, slightly demented, very mildly demented, and moderately demented. ResNet50 outperforms InceptionV3, achieving an accuracy of 93.9%, a micro F1 score of 94%, and a macro F1 score of 96%, demonstrating its efficacy and consistent performance in detecting and classifying all categories. Compared with current models, the suggested technique is more effective.