Exploring 3D Convolutional Neural Network Models for Alzheimer’s Disease Classification Based on 3D MRI Images

Abstract

Alzheimer’s disease is a common form of progressive dementia, especially among the elderly, and is characterized by a decline in cognitive function. Classifying this disease using 3D brain imaging through MRI is challenging due to the complexity of the data and the similarity of features across classes. This study develops a classification model based on a 3D Convolutional Neural Network (3D CNN) architecture, specifically using ResNet-18. The dataset used is obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), consisting of 1,281 samples evenly distributed across three classes: Cognitively Normal (CN), Mild Cognitive Impairment (MCI), and Alzheimer’s Disease (AD). The data undergo several preprocessing steps, including skull stripping, normalization, and augmentation. The model is tested in two configurations: without dropout and with a dropout rate of 0.3. The results show that the model with dropout performs better, achieving a classification accuracy of 62.0% and a macro F1-score of 0.604. The model outperforms ADNet and Vision Transformer, and approaches the accuracy of Vision Mamba. Nevertheless, this approach still requires further development, particularly in improving accuracy for the CN class and reducing performance imbalance across classes.