Deep Learning-Based Classification of Alzheimer’s Disease Using MRI Scans: A Customized Convolutional Neural Network Approach

• Published in: SN computer science Vol. 5; no. 7; p. 917
• Main Author: Shastry, K. Aditya
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 30.09.2024; Springer Nature B.V
• Subjects: Accuracy; Alzheimer's disease; Artificial neural networks; Brain research; Business metrics; Caregivers; Classification; Cognition; Cognitive ability; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Data Structures and Information Theory; Datasets; Deep learning; Disease; Effectiveness; Information Systems and Communication Service; Machine learning; Magnetic resonance imaging; Medical imaging; Medical research; Methods; Neural networks; Original Research; Pattern Recognition and Graphics; Public health; Quality of life; Research Advancements in Intelligent Computing; Software Engineering/Programming and Operating Systems; Vision; Deep learning; Convolution Neural Network; Alzheimer’s disease; Magnetic resonance imaging; Classification
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-024-03284-4

Alzheimer's Disease (AD) is a complex condition impacting memory, cognition, and daily functions. Early diagnosis is crucial for effective management and treatment, although no cure currently exists. In this study, we develop and evaluate a custom Convolutional Neural Network (c-CNN) model for AD classification using MRI scans. Our c-CNN model, featuring a unique architecture with two convolution layers, one max pooling layer, seven sequential layers, two dropout layers, and one dense output layer, is tested on three distinct MRI datasets. The first dataset comprises 6400 images categorized into four classes: Mild Demented (MiD), Moderate Demented (MdD), Non-Demented (ND), and Very Mild Demented (VMD). The second dataset includes 1296 images representing five stages of AD from the AD Neuroimaging Initiative (ADNI) repository. The third dataset contains 5154 images across three classes: AD, Confidence Interval (CI), and cognitive normal (CN). We rigorously compare our c-CNN model against DenseNet-121 (D-121), InceptionV3 (IV3), and Resnet 50 (Res-50) on these datasets. Our model outperforms others in accuracy, precision, recall, F1-score, area under the curve (AUC), and validation loss. These results underscore the effectiveness of our c-CNN model, highlighting its potential as a valuable tool for accurate AD classification.