Pathological and neuroimaging alterations in the cerebellum of Alzheimer's disease (AD) patients have been documented. However, the role of cerebellum-derived radiomic and structural connectome modeling in the prediction of AD progression remains unclear. Radiomic features were extracted from magnetic resonance imaging (MRI) in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset (n=1319) and an in-house dataset (n=308). Integrated machine learning models were developed to predict the conversion risk of normal cognition (NC) to mild cognitive impairment (MCI) over a 6-year follow-up. The cerebellar models outperformed hippocampal models in distinguishing MCI from NC and in predicting transitions from NC to MCI across both cohorts. Key predictors included textural features in the right III and left I and II lobules, and network properties in Vermis I and II, which were associated with cognitive decline in AD. Cerebellum-derived radiomic-network modeling shows promise as a tool for early identification and prediction of disease progression during the preclinical stage of AD. Altered cerebellar radiomic features and topological networks were identified in the subjects with mild cognitive impairment (MCI). The cerebellar radiomic-network integrated models outperformed hippocampal models in distinguishing MCI from normal cognition. The cerebellar radiomic model effectively predicts MCI risk and can stratify individuals into distinct risk categories. Specific cerebellar radiomic features are associated with cognitive impairment across various stages of amyloid beta and tau pathology.
Read full abstract