Abstract:
Alzheimer’s disease is a widespread neurological disorder which affects millions of people around the world and presents complex challenges for patients, families, and healthcare providers. A mix of non-pharmaceutical and pharmacological strategies are needed to manage AD in order to reduce symptoms and enhance quality of life during various phases of the disease. Since the course of AD varies greatly, personalized treatment strategies arevital, with early detection and stage-specific classification playing a key role. Medical imaging, particularly MRI, has become a cornerstone in diagnosing and tracking AD, though its interpretation can be slow and heavily dependent on clinical expertise. To address this issue, we propose a customized CNN model, called ALSA-3, which leverages deep learning for MRI-based AD classification. To refine image quality, methods such asPSNR, RMSE, MSE, and SSIM are applied, while an ablation study fine-tunes the architecture by adjusting hyperparameters and layer configurations. Experimental results demonstrate that ALSA-3 outperforms conventional approaches, reaching an impressive 99.50% accuracy along with precision, recall, and F1-scores of 100%, 99%, and 99% respectively. The model’s robustness was further validated across different k-fold values, confirming its reliability. In addition, explainable AI techniques like Grad-CAM and Grad- CAM++ were integrated to make the decision-making process more transparent and interpretable for clinicians. All things considered, this study advances the field of AD categorization research and provides useful insights for patients, caregivers, and the medical community. ALSA-3 has the potential to provide quicker and more accurate diagnoses, which would eventually lead to improved treatment for people with AD because of its high accuracy and efficiency.
