Comparative Classification of Alzheimer’s Disease

Abstract

Alzheimer’s disease (AD) is recognized as a neurodegenerative condition characterized by the progressive decline of memory, cognitive abilities, and higher brain functions. Rather than constituting a singular ailment, AD represents a cluster of related disorders sharing common traits. Employing pattern classification methods rooted in deep neural networks, such as convolutional neural networks (CNNs), facilitates the categorization of patients into distinct AD subtypes and the differentiation of various stages of disease severity. This study concentrates on the early identification of diverse phases of cognitive aging and AD leveraging neuroimaging and transfer learning (TL). Magnetic resonance imaging (MRI) scans sourced from the Alzheimer’s Dataset on Kaggle, encompassing several classes including non-dementia (NONDEM), very mild dementia (VERDEM), mild dementia (MILDEM), and moderate dementia (MODDEM), are subject to classification using a transfer learning framework. Specifically, the research assesses the classification efficacy of three pre-trained networks-VGG-19, ResNet-50, and InceptionV3. These networks are trained and evaluated on a dataset comprising 6400 images from the Kaggle repository. The classification performance of these models is evaluated through the utilization of confusion matrices and associated metrics. Notably, VGG-19, ResNet-50, and Inception-V3 demonstrated overall accuracies of $92.86 \%, 85.99 \%$, and $91.04 \%$, respectively, in detecting AD.