Explainable Alzheimer's Disease Classification from Brain MRI using an Attention-Enhanced Residual Network
Document Type
Conference Proceeding
Publication Date
2026
Abstract
Alzheimer's Disease (AD) is the most common cause of dementia worldwide. The condition has had significant implications for healthcare. While automated structural MRI staging is crucial for early intervention in Alzheimer's Disease, many deep learning models have been criticized for achieving superior performance at the cost of interpretability. This paper presents an explainable Alzheimer's Disease classification model from brain MRI images using an Attention-Enhanced Residual Network in three stages. The first stage involves developing an extensive preprocessing pipeline for standardizing input images. The second stage involves developing HybridResNet, an attention-enhanced transfer learning model that utilizes ResNet-50 and a convolutional block attention module. This stage also involves developing a classification system that utilizes Focal Loss and a two-step tuning approach for handling class imbalance. The third stage involves developing a system that utilizes Gradient-Weighted Class Activation Mapping (Grad-CAM) for Explainable artificial intelligence (XAI). The system was tested using the Augmented Alzheimer MRI Dataset, consisting of 33,984 images. The HybridResNet model achieved 99.66% precision, 99.67% F1-score, and an Area Under Curve (AUC) score of 1.000. The system focuses on key areas that are crucial for diagnosis, making it reliable.
Recommended Citation
Fouad, H. A., & Sayed, G. I. (2026, July). Explainable Alzheimer's Disease Classification from Brain MRI using an Attention-Enhanced Residual Network. In 2026 Intelligent Methods, Systems, and Applications (IMSA) (pp. 300-305). IEEE. https://doi.org/10.1109/IMSA70415.2026.11700260