An Automated Framework and Medical Intelligent System for MRI-Based Alzheimer's Diagnosis Classification.
Authors
Affiliations (1)
Affiliations (1)
- College of Technological Innovation, Zayed University, Abu Dhabi, United Arab Emirates. [email protected].
Abstract
In the era of artificial intelligence (AI), Alzheimer's disease (AD) can be diagnosed through magnetic resonance imaging (MRI) at accurate times and with precision to make effective clinical interventions. Nevertheless, classification is still problematic because of low-contrast anatomical structures, inter-class variations, imbalances of the classes, and the possibility of data leakage in slice-based methods. To overcome these limitations, this paper will introduce a contrast-sensitive hybrid Swin Transformer V2 that employs a 3D Convolutional Neural Network that is trained with Harris Hawks Optimization (CA-Swin3DNet-HHO) to classify AD using MRI. The framework presented includes gamma-logarithmic contrast enhancement, skull stripping, and intensity normalization to enhance the visibility of disease-relevant features. A 3D-CNN module is used to encode volumetric structural features, and Swin Transformer V2 is trained to learn global contextual dependencies based on efficient window-based self-attention. Automatic hyperparameter tuning is done using Harris Hawks Optimization (HHO), which enhances convergence and generalization. A feature fusion approach combines both local and global features to improve the classification performance in cognitively normal (CN), mild cognitive impairment (MCI), and AD classes. Strict subject-level partitioning is used to evaluate the model on the ADNI-2 and OASIS-2 datasets to avoid data leakage. The experimental results show better performance with 99.3% accuracy, 99.1% F1-score, and 0.996 AUC on ADNI-2 and 98.7% accuracy, 98.3% F1-score, and 0.991 AUC on OASIS-2. Also, Grad-CAM-based visualization can emphasize clinically relevant parts of the brain, enhancing the interpretability and reliability of the model. In general, the suggested framework offers a precise, solid, and readable solution to automated diagnosis of Alzheimer's disease and has a high potential for clinical use in the real world.