Benchmarking Convolutional Neural Networks for Alzheimer’s Disease Diagnosis

Abstract

Alzheimer’s disease (AD), a major form of dementia, can have its symptoms managed more effectively when detected in the early stages. In recent years, computer-aided diagnostic (CAD) systems using magnetic resonance imaging (MRI) have demonstrated promising results in the classification of AD. In this study, T1-weighted MRI images are processed using FreeSurfer software to extract cortical and subcortical features. Data from 190 subjects obtained from the ADNI database are used for analysis.
To reduce model complexity, simplified architectures with a single layer from well-known convolutional neural network (CNN) models—ResNet, VGG, and AlexNet—are employed. A multi-class classification approach is applied to categorize subjects into four groups: cognitively normal (CN), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and Alzheimer’s disease (AD). Experimental results show that the VGG model achieved the highest classification accuracy of 96%, followed by GoogLeNet (93%), ResNet (91%), and AlexNet (89%).