Characterizing heterogeneous atrophy patterns of Alzheimer's disease dementia with mixture‐of‐experts (MOE) for joint classification and clustering

This study demonstrated the effectiveness of MOE model in both classification and characterization of the AD heterogeneity. By approximating non-linear decision boundaries, the MOE model enhances the classification performance while simultaneously providing interpretable subtype information. This method offers valuable insights into the automatic and interpretable diagnosis of Alzheimer's disease and potentially other neurodegenerative diseases.