The paper presents a framework which synergistically integrates MRI and PET imaging modalities to enhance diagnostic accuracy for Alzheimer’s Disease (AD). Leveraging the detailed anatomical imaging of MRI along with the metabolic information from PET, the framework utilizes sophisticated data pre-processing methods such as intensity normalization and non-local means filtering to guarantee high-quality input data and reduce variability among subjects. A significant advancement is the modification of the 3D U-Net architecture which features several skip connections for accurate segmentation of the hippocampus. The segmentation procedure is further optimized through a hybrid algorithm that combines the Skill Optimization Algorithm with the Deep Sleep Optimizer, leading to improved convergence and performance. The integrated loss function, which includes a weighted cross-entropy and the Dice loss coefficient, markedly enhances spatial overlap accuracy in segmentation tasks. The Wide Context mechanism enhances the network’s capability to capture broad contextual information, which is essential for intricate delineation of the hippocampus. Feature extraction involves a varied set of attributes, ensuring a thorough representation of both the morphology and intensity traits of the hippocampus. The detection network, which incorporates advanced components like 3D CNNs, ZFNet, RNNs, and attention mechanisms, enables the identification of spatial and temporal patterns in multimodal data. This approach lays a robust groundwork for future research and therapeutic applications, representing a significant advancement in the early detection of hippocampal anomalies associated with Alzheimer’s disease.