ALZENET: Deep learning-based early prediction of Alzheimer's disease through magnetic resonance imaging analysis

Abstract

Alzheimer's Disease (AD) is a particular form of dementiacharacterized by a steady decline in brain function and cognitive abilities, leading to memory loss, disorientation, and trouble doing daily tasks. It generally occurs in people aged over 65, and currently, there are around 55 million Alzheimer's patients which is becoming a growing concern, and the death rate due to AD has increased significantly. Existing ways of categorizing patients include medical records, cognitive testing, and magnetic resonance imaging (MRI), which assist in diagnosing the condition, but a deep learning-based approach helps to discover Alzheimer's at an early stage. In this study, we propose an Alzheimer Recognition Ensemble Network (ALZENET) for classifying various stages of Alzheimer using MRI data. The study utilized the Kaggle MRI-based AD dataset, which comprised a total of 6,400 MRI images across four distinct classes. To address the significant class imbalance present in the dataset, the SMOTE technique was employed. Our ensemble model, which combines VGG16, Inception V3, and ResNet50 with a multi-layered CNN architecture, achieved a test accuracy of 97.31 %, significantly outperforming existing approaches in Alzheimer's disease classification. Additionally, a deep learning-based web application was developed to assist doctors in efficiently identifying different stages of Alzheimer's disease. This approach provides a promising tool for the early and accurate detection of Alzheimer's disease, potentially improving patient outcomes.