Alzheimer’s disease stage prediction using a novel transfer learning-Alzheimer’s network architecture

Abstract

The root cause of Alzheimer’s disease (AD) is unknown except for a very tiny number of family instances caused by a genetic mutation. A thorough examination of particular brain disorders’ tissues is necessary to correctly identify the circumstances using scans of magnetic resonance imaging (MRI), and specific non-brain tissues, like the neck, skin, muscle, and fat, make further investigation challenging and can be seen in MRI scans. This work aims to use the FSL-BET skull stripping tool to remove non-brain tissues and extract the significant region of the brain- deep learning (DL) techniques rather than machine learning (ML) models helpful in classification and predictions. The most frequent issue with DL models is which needs a lot of training data, causes to problems with class imbalance. To avoid imbalance issues, we used data augmentation to ensure that the samples were distributed equally among the classes. A novel transfer learning Alzheimer’s disease network (TL-AzNet) based visual geometry group-19 (VGG19) technique was developed in this study. Conducted a comparison study using the base and suggested models, comparing over data with oversampling versus non-oversampling. The novel model predicted AD with a 95% accuracy rate.