Abstract
OBJECTIVE: Brain age gap estimation (BrainAGE) has demonstrated accelerated brain aging in mild cognitive impairment (MCI) and functional aging in patients with Alzheimer's disease (AD). Nevertheless, the neuroanatomical aging characteristics of AD remain insufficiently understood. The present study aimed to investigate the neuroanatomical aging conditions of AD using the BrainAGE model. METHODS: Clinical profiles and T1 structural magnetic resonance imaging (MRI) data of 219 healthy controls (HCs) and 51 AD patients were collected. We extracted gray matter and white matter volumes from the structural MRI and used the BrainAGE model to evaluate aging characteristics in AD patients. Specifically, we configured a stacking model with two levels to predict brain age. The model was trained on the 219 HCs and tested on the AD patients to investigate whether AD could lead to different neuroanatomical aging conditions. In addition, we explored differences in voxel-wise gray matter, white matter patterns, and clinical profiles between AD patients with different neuroanatomical aging conditions. RESULTS: The proposed machine learning algorithm could accurately estimate brain age in HCs. Application of the BrainAGE model to the AD group revealed three subgroups with advanced, typical, and delayed brain aging conditions. The three AD subgroups also differed in voxel-wise gray matter and white matter volumes. Furthermore, the three subgroups differed in age and genetic scores. CONCLUSION: The BrainAGE model identified subtle deviations from age-related brain atrophy in the AD cohort with distinctive clinical manifestations, which contribute to the understanding of neuropathology of AD.