An improved strategy for predicting Alzheimer's disease progression in patients with mild cognitive impairment by clustering hippocampal shape data

Alzheimer's disease (AD), a neurodegenerative disorder leading to dementia and cognitive deficits, is influenced by various risk factors. Precise prognoses of AD are crucial to developing effective treatments, and recognizing potential heterogeneity within the population aids prognoses by detecting possible risk factors of each subgroup. Hippo cam- pal structure changes have been connected to Alzheimer's disease and are hypothesized to occur early in the illness's development. In our work, we propose a universal clustering method for metric spaces that combines energy distance and a decision tree to detect the underlying heterogeneity in hippocampal structure. We are able to show that our method helps highlight the importance of utilizing hippo campus information for diagnosis and treatment and improves the prediction accuracy in a case study. The effectiveness of our approach is further backed up by the conducted simulations on widely-used metric spaces.