Ensemble deep learning for Alzheimer’s disease characterization and estimation

Abstract

Alzheimer’s disease, which is characterized by a continual deterioration of cognitive abilities in older people, is the most common form of dementia. Neuroimaging data, for example, from magnetic resonance imaging and positron emission tomography, enable identification of the structural and functional changes caused by Alzheimer’s disease in the brain. Diagnosing Alzheimer’s disease is critical in medical settings, as it supports early intervention and treatment planning and contributes to expanding our knowledge of the dynamics of Alzheimer’s disease in the brain. Lately, ensemble deep learning has become popular for enhancing the performance and reliability of Alzheimer’s disease diagnosis. These models combine several deep neural networks to increase a prediction’s robustness. Here we revisit key developments of ensemble deep learning, connecting its design—the type of ensemble, its heterogeneity and data modalities—with its application to AD diagnosis using neuroimaging and genetic data. Trends and challenges are discussed thoroughly to assess where our knowledge in this area stands. In this Review, the authors cover the latest understanding of ensemble deep learning models as a means to complement Alzheimer’s disease diagnosis.