Asymmetric degeneration of the hippocampus predicts risk of Alzheimer’s disease

Abstract

AbstractBackgroundAsymmetry of the hippocampus is an important finding in normal aging and Alzheimer disease (AD). However, it is largely unknown how asymmetric degeneration of the hippocampus is associated with an individual's risk of AD.MethodA deep‐learning (DL) classification model was developed to quantify degeneration of the hippocampus based on its magnetic resonance imaging (MRI) scan and binary mask. Different from most extant studies, the DL model was trained to distinguish AD patients from cognitively normal (CN) controls with their bilateral hippocampi as two separate instances for generating two classification scores (positive indicating AD‐like and negative indicating CN‐like), one for each hippocampus. An integration of the two scores of an individual’s MRI data was utilized to classify AD patients and CN controls, to predict individual mild cognitive impairment (MCI) subjects’ progression to AD dementia in a time‐to‐event analysis setting, as well as to identify subgroups of MCI subjects with distinct risk of AD dementia. Particularly, the DL classification model was trained on the ADNI 1 cohort (n=420), and then validated for distinguishing AD subjects from CN controls on the ADNI 2&GO (n=469) and AIBL (n=396) cohorts and for predicting MCI subjects’ progression to AD dementia and identifying MCI subgroups on the ADNI 2&GO cohort (n=439).ResultThe DL model achieved accurate classification performance with classification rates of 0.923 and 0.927 and AUC values of 0.963 and 0.961 on the ADNI‐GO&2 and AIBL cohorts, respectively. The integration of the bilateral hippocampal classification scores achieved a concordance index of 0.765 for predicting individual MCI subjects’ progression to AD dementia. The MCI subgroups with bilateral AD‐like, bilateral CN‐like, and one AD‐like and one CN‐like scores were significantly different in risk of progression to AD dementia and a variety of biomarkers and cognitive scores, including Apoe4, Amyloid, tau, and ADAS‐cog score (p < 0.0004).ConclusionThe present study demonstrated that the asymmetric hippocampal degeneration patterns quantified using a DL model on MRI data could achieve promising classification performance for distinguishing AD patients from CN controls and may aid the clinicians to early prediction of MCI subjects’ progression to AD dementia to facilitate personalized medicine.