Alzheimer’s Disease Dementia Detection Using Transfer Learning Based Convolutional Neural Network Model

Alzheimer’s disease (AD) is the most common type of dementia that is not a result of natural aging. In most countries, it is one of the leading causes of death for seniors. One of the main methods for detecting AD is by magnetic resonance imaging (MRI). Several machine learning (ML) and deep learning (DL) techniques have been put out in recent years for the automated detection of AD, but the accuracy of these techniques is currently limited. So, the primary goal of this research is to offer a model for precise AD detection that is based on transfer learning. For the purpose of detecting AD in this paper, we have used two different models. We studied and evaluated models from two different platforms, GoogleNet and residual networks (ResNet). Different variants of ResNet (ResNet18, ResNet34, ResNet50, and ResNet101) were studied. In this study, specificity, accuracy, positive predictive rate, sensitivity, F1 score, balanced accuracy, Fowlkes-Mallows index, and Youden’s J statistics were investigated. ResNet18, ResNet34, ResNet50, and ResNet101 have shown a precision of 97.28%, 98.25%, 98.41%, and 98.57%. GoogleNet has accuracy rate of 96.32%. The learning curve is also presented in this work, and shows a good fit for the ResNet101 model. The proposed networks were compared on the basis of computational time. ResNet101 performed better than other networks in all the parameters and had the largest computational time. MRI images from the Alzheimer’s disease neuroimaging initiative (ADNI) database have been used and compared with similar work based on ML and DL. A comparison with the existing methods showed that the proposed method could help in the reliable detection of AD.