A Vertical Stacking-Based Ensemble Deep Learning Model for Early Diagnosis of Alzheimer's Disease Using Multimodal MRI Scans

Early detection of Alzheimer's Disease (AD) is crucial for timely interventions that can slow disease progression, enhance quality of life, and assist with future planning. Convolutional Neural Networks (CNNs) are an efficient method for processing image-based data. In this work, we used CNN-based deep learning models to extract structural information from structural MRI (sMRI) and brain neuron connectivity patterns from functional MRI (fMRI) data. In this study, a stacking-based ensemble multimodal framework was proposed by integrating both texture features and brain neuron connectivity patterns using Deep Learning (DL) models such as GoogLeNet, DenseNet-121, GNN, and U-Net. The prediction probabilities were combined using a vertical stacking approach to create ameta-feature matrix, which was utilized by the Meta model and trained using the Random Forest classification algorithm to generate the final predictions. This approach leveraged the complementary strengths of structural and functional data, thereby improving classification accuracy and generalization. The proposed method demonstrated remarkable accuracy of95.18%, reflecting its exceptional performance and minimal error rates. It surpassed the effectiveness of existing state-of-the-art methods, showing high precision in early AD detection and highlighting its potential for neurodegenerative disease research.