Classification of Alzheimer's diseases from PET Images Using a Convolutional Neural Network

Alzheimer's disease, a neurodegenerative and incurable illness, poses a significant challenge. Early and accurate diagnosis, achieved through non-invasive methods, paves the way for timely intervention and offers hope in managing this complex disease. Our research introduces a deep learning method for classifying Alzheimer's disease stages. Instead of relying on complex pre-trained networks, we designed a streamlined 2D Convolutional Neural Network (CNN) architecture specifically for analyzing PET brain images. To display the effectiveness of our approach, we employed two distinct CNN architectures: one with two convolutional layers and another with five convolutional layers. This study utilized data from the ADNI (Alzheimer's Disease Neuroimaging Initiative) public database. The dataset represents three disease classes: CN (cognitively normal), MCI (mild cognitive impairment), and AD (Alzheimer's disease). The data was split with 80% allocated for training the model and 20% reserved for testing its performance. In addition, we compared our approach to other deep learning architectures, such as ResNet50 and VGG16. The reported results confirmed the high accuracy and resilience of the proposed method, as evidenced by the overall test accuracy of 99.60%.