<![CDATA[AbstractGeneral Background: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder requiring accurate and accessible diagnostic support. Specific Background: Magnetic Resonance Imaging (MRI) is widely used for structural brain assessment, and Convolutional Neural Networks (CNN) enable automated feature extraction from medical images. Knowledge Gap: Prior studies report high classification performance but rarely integrate comprehensive evaluation with real-time deployment for decision support. Aims: This study develops and evaluates a CNN-based model for classifying 2D axial MRI images into Alzheimer’s Disease (AD), Mild Cognitive Impairment (MCI), and Common Normal (CN), alongside web-based implementation. Results: Using approximately 5,000 ADNI MRI images, the model achieved 97.67% accuracy, 97.73% precision, 97.67% recall, and 97.65% F1-score, with AUC values near 1.00. Learning curves indicated stable convergence without overfitting or underfitting, and confusion matrix analysis confirmed consistent multi-class discrimination. The deployed Hugging Face–Gradio application generated predictions in under five seconds per scan without performance degradation. Novelty: This research combines rigorous multi-metric validation with interactive web deployment as an artificial intelligence decision support system for early AD screening. Implications: The findings demonstrate the technical feasibility of CNN-based MRI classification for preliminary cognitive disorder screening, while emphasizing the need for multimodal integration and prospective clinical validation. Highlights• Achieved robust multi-class discrimination among AD, MCI, and CN categories using axial brain scans.• Demonstrated stable training dynamics validated through loss convergence and receiver operating characteristics.• Implemented an interactive artificial intelligence platform with sub-five-second prediction time. KeywordsAlzheimers Disease; Convolutional Neural Networks; MRI Classification; Deep Learning; Clinical Decision Support System]]>
Copyrights © 2026
