Classification of Alzheimer’s Disease by Magnetic Resonance Imaging with an Ensemble Approach

Alzheimer’s Disease (AD), affecting over 55 million people globally, demands reliable diagnostic tools. Single-model approaches using CNNs and traditional ML face critical limitations. This study proposes two frameworks: a stacking-CNN ensemble (VGG-16, ResNet-101, DenseNet-121) and two voting ML ensembles (Voting[all]: KNN, RF, SVC, LR, XGBoost; Voting[few]: KNN, RF, XGBoost). Evaluated on 6,400 MRIs, Voting[few] achieved the highest classification metrics (97.8% accuracy; 0.984 MCC; 93.8% F1macro), outperforming individual CNNs, validated through Friedman-Nemenyi tests. Results suggest, in this context, that simpler ML models might better capture the inherent characteristics of MRI data for AD diagnosis.