Four-Stage Alzheimer's Disease Classification from MRI Using Topological Feature Extraction, Feature Selection, and Ensemble Learning
This study addresses the challenge of accurately classifying Alzheimer’s disease (AD) across its four stages—from non-demented to moderate—under conditions of limited MRI data and stringent requirements for model interpretability. To this end, it proposes a novel approach that integrates topological data analysis (TDA) with ensemble learning, extracting topological features from brain structures and incorporating feature selection strategies. Notably, the method achieves high classification performance without relying on data augmentation or pre-trained models. Evaluated on the OASIS-1 dataset, the proposed model attains an accuracy of 98.19% and an AUC of 99.75%, matching or surpassing current deep learning-based approaches while maintaining low computational overhead and offering strong clinical interpretability.