Multi-Stages of Alzheimer's Disease Classification Using Deep Learning

Abstract

Alzheimer's disease (AD) is a genetic disorder that is characterized by a gradual deterioration in cognitive function and is the most prevalent cause of dementia. Early and precise identification of AD stages is essential for efficient therapy and control. The existing diagnostic techniques, which mainly depend on Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) scans, are inefficient and susceptible to mistakes made by humans, resulting in delays in diagnosing and treating patients. This study aims to propose a deep learning model for classifying the different stages of AD using scans. This study evaluates the performance of three deep learning models, including Convolutional Neural Networks (CNN), ResNet50 and VGG16, on the MRI dataset collected from Kaggle, consists of images classified into four categories: no dementia, very mild dementia, mild dementia, and moderate dementia. Our results show that CNN outperformed the other models with an accuracy of 88.4%, demonstrating high sensitivity and specificity across different stages. The system achieved an accuracy of 97% in classifying AD into multiple categories. CNN exhibited strong performance in accurately differentiating between various stages of AD, as indicated by the results of the receiver operating characteristic (ROC) curve. Future research will prioritize the integration of supplementary clinical datasets to enhance the resilience of the model and broaden its diagnostic capabilities to encompass other neurodegenerative disorders.