Comparative Analysis of Convolutional Neural Network and U-Net Segmentation Model for Alzheimer’s Disease Detection

Alzheimer’s Disease (AD) is a progressive neurodegenerative disorder that demands the deployment of precise, automated, and reproducible diagnostic methodologies for effective clinical management and therapeutic decision-making. This study proposes and comparatively evaluates two computational frameworks for the detection of AD utilizing structural Magnetic Resonance Imaging (MRI) data. The first framework integrates overlay-based image segmentation techniques, employing intensity thresholding and pixel-wise differentiation, followed by classification of the extracted regions using a Convolutional Neural Network (CNN) architecture. The second framework incorporates a U-Net-based semantic segmentation model, coupled with ensemble classification schemes comprising Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbors (KNN) algorithms. A comprehensive quantitative analysis is performed to assess segmentation accuracy metrics and classification performance indices, including precision, recall, F1-score, and overall classification accuracy. Additionally, the influence of optimization algorithms-specifically Adam and RMSProp-on the convergence behavior and classification efficacy of the CNN model is systematically investigated. The proposed methodologies demonstrate classification accuracies within the range of 70% to 89%, providing comparative insights into the efficacy of conventional and deep learning-based segmentation-classification pipelines. The findings contribute to the advancement of neuroimaging-based diagnostic systems and offer critical guidance for researchers and clinicians in the selection of optimal computational approaches for medical image analysis applications.