<p>Everyone has noted the great progress registered in the medicine sector. Many formerly complicated diseases are solved. Among a lengthy list, one can name, e.g., the one named Alzheimer’s disease (AD). This worldwide illness remains a major public health concern. An early and an accurate stage diagnosis is critical for effective intervention. Many solutions were suggested but unfortunately This sickness is simply another significant human obstacle that needs to be overcome. In this context, our paper suggests being helped by recent computing tools to solve this challenge. Deep Learning (DL) has advanced AD diagnosis. This tool enhances formerly limited interpretability, class imbalance, and the “black box” nature of models hinder clinical adoption. For that, we propose XCAD-AD, an explainable computer-aided diagnosis (CAD) system. This approach integrates DL for automated hippocampal segmentation and multi-stage AD classification. Segmentation is performed using the encoder–decoder LinkNet architecture. In addition, the classification leverages pre-trained convolutional neural networks (CNNs) with deep transfer learning (VGG19, ResNet50, and Inception v3) alongside a custom three-dimensional CNN (3D CNN). To improve transparency and clinical trust, Gradient-weighted Class Activation Mapping (Grad-CAM) is used to highlight decision-relevant brain regions. Evaluated on a publicly available benchmark dataset, the proposed 3D CNN achieved the best results. In fact, the following outputs were obtained in terms of accuracy: 99%, 98.27% precision, 96.85% recall, and 99.23% F1-score. This proves high robustness in distinguishing AD stages. The LinkNet model achieved highly accurate hippocampal segmentation with a Dice metric of 0.9938, outperforming existing approaches. These findings attest to our XCAD-AD as a robust, interpretable, and fully automated pipeline for AD diagnosis. This offers strong potential for integration into clinical workflows.</p>