A Weighted Ensemble Deep Learning Approach for Five-Class Alzheimer’s Disease Classification from DICOM MRI Images
Applied Sciences (Switzerland), cilt.16, sa.11, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 16 Sayı: 11
- Basım Tarihi: 2026
- Doi Numarası: 10.3390/app16115466
- Dergi Adı: Applied Sciences (Switzerland)
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, INSPEC, Directory of Open Access Journals
- Anahtar Kelimeler: Alzheimer’s disease, classification, deep learning, MRI
- Gazi Üniversitesi Adresli: Evet
Özet
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and memory impairment, most commonly observed in older adults. Accurate classification of AD-related conditions from magnetic resonance imaging (MRI) plays an important role in supporting research and analytical studies in medical imaging. In this study, we investigate the performance of deep learning models for multi-class classification of AD-related diagnostic categories using MRI data. Convolutional Neural Network (CNN), Residual Network (ResNet-50), and Densely Connected Convolutional Network (DenseNet121) architectures were evaluated on a five-class dataset derived from the TR-MoH (Ministry of Health of the Republic of Türkiye), consisting of Alzheimer’s disease–related diagnostic codes. A consistent preprocessing pipeline, including slice extraction, resizing, normalization, and data augmentation, was applied prior to model training. In addition, experiments on a publicly available Kaggle dataset were conducted to assess model behavior across datasets with different characteristics. Grad-CAM visualizations were additionally employed to improve model interpretability by highlighting the brain regions that contributed most to the classification decisions. The results show that individual models achieved accuracy values ranging from 89.31% to 91.07% on the TR-MoH dataset and from 98.86% to 99.93% on the Kaggle dataset. A weighted ensemble approach combining multiple architectures yielded the most effective results, reaching 93.05% and 99.94% on the respective datasets. These results indicate that deep learning models can effectively learn discriminative patterns from heterogeneous MRI data and perform multi-class classification tasks with notable class differentiation capability. However, the results should be interpreted within the scope of the defined classification problem and dataset characteristics. Overall, the study highlights the potential of ensemble-based deep learning approaches for supporting MRI-based categorization of AD-related conditions in a research context.