Deep ensemble learning approach for multi-stage diagnosis of Alzheimer’s disease

Abstract Alzheimer’s disease (AD), which is one of the leading causes of dementia, is a degenerative neurological condition that typically develops slowly and worsens over time. Alzheimer’s disease affects memory and cognitive abilities, and it is a prevalent condition worldwide. Early diagnosis of the disease is crucial to mitigate its irreversible effects, but traditional diagnostic methods have limitations. Recent advancements in neuroimaging, biomarkers, and deep learning algorithms have opened up new possibilities for early detection and accurate diagnosis of the condition. In this research, we develop a multi-stage ensemble classifier model that can classify Alzheimer’s disease into four categories: very mild dementia, mild dementia, moderately severe dementia, and non-dementia, using 6400 MRI scans. To address the black-box nature of deep learning models in medical imaging, we integrate Grad-CAM visualization techniques to interpret the classification results of Alzheimer’s disease stages. This method allows us to identify the most influential regions in MRI scans that drive the model’s decisions, thereby increasing the reliability and clinical trust in the automated diagnosis process. The proposed deep ensemble model achieved an accuracy of 98.71% and an AUC of 99.13%, outperforming other state-of-the-art methods.