ADD-Net-MRI: An Explainable Deep Learning Approach for Early Alzheimer’s Detection

Early and accurate identification of Alzheimer’s disorder (AD) is crucial for powerful medical intervention, mainly in the course of the slight cognitive impairment (MCI) level. Structural magnetic resonance imaging (MRI) offers precious neuroanatomical records; but automated analysis remains tough because of elegance imbalance, inter-problem variability, confined generalization, and inadequate explainability. In these paintings, we suggest ADD-Net-MRI, a generalizable and explainable deep mastering framework for early and multi-level Alzheimer’s disorder class using structural MRI. The framework employs a light-weight convolutional neural community trained from scratch to stability diagnostic performance and computational efficiency. To deal with elegance imbalance, concern-degree SMOTETOMEK sampling is integrated throughout schooling, at the same time as a standardized preprocessing pipeline ensures reproducibility and robustness throughout datasets. Model interpretability is carried out using Grad-CAM– primarily based visual factors, permitting clinically significant insight into sickness-associated mind regions. Extensive problem-level pass-domain experiments demonstrate that ADD-Net-MRI achieves competitive class performance, improved sensitivity for early-degree AD detection, and stronger explainability with reduced computational complexity. These outcomes advocate that ADD-Net-MRI can function an effective laptop-aided diagnostic tool to support early Alzheimer’s disease assessment.