Enhancing Alzheimer's Disease Detection Using Advanced MRI Preprocessing and Deep Learning Techniques

This study investigates how optimized Magnetic Resonance Imaging (MRI) preprocessing enhances the performance of Artificial Intelligence (AI) models for Alzheimer's Disease (AD) detection. A standardized preprocessing pipeline, comprising noise reduction, skull stripping, and intensity normalization, was integrated into a hybrid Deep Learning (DL) model called MRI-Net. MRI-Net combines Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and Vision Transformers (ViTs) to leverage complementary spatial and contextual information. Using the publicly available ADNI dataset, MRI-Net achieved a classification accuracy of 94.7%, surpassing baseline CNN (88.2%), CNN+RNN (90.5%), and ViT (92.8%) models. The inclusion of a structured preprocessing pipeline improved anatomical clarity, reduced inter-subject variability, and enhanced feature discriminability. These findings demonstrate that optimized preprocessing substantially strengthens model generalization and diagnostic reliability. The proposed MRI-Net framework highlights the importance of integrating preprocessing and hybrid AI architectures to support early-stage AD detection and clinical decision-making.