HAI Book 2025 - Flipbook - Page 326
He, Hengda
67
Network-based amyloid-β pathology predicts subsequent cognitive
decline in cognitively normal older adults
Hengda He1, Qolamreza Razlighi2, Yunglin Gazes3, Christian Habeck1,4,5, Yaakov Stern1,4,5
1
Cognitive Neuroscience Division, Department of Neurology, Columbia University, Vagelos College of Physicians and
Surgeons, New York, NY, US
2
Quantitative Neuroimaging Laboratory, Brain Health Imaging Institute, Department of Radiology, Weill Cornell
Medicine, New York, NY, US
3
Center for Biomedical Imaging & Neuromodulation, Nathan S. Kline Institute for Psychiatric Research, Orangeburg,
NY, US
4
Taub Institute for Research in Alzheimer’s Disease and the Aging Brain, Columbia University, New York, NY, US
5
Gertrude H. Sergievsky Center, Columbia University, New York, NY, US
Background: The deposition of amyloid-´ protein in the human brain is a hallmark of Alzheimer9s disease and is
often related to cognitive decline. However, the relationship between early amyloid-´ deposition and future
cognitive impairment remains poorly understood, particularly concerning its network-level effects. Here, we
employ a cross-validated machine learning approach, and aim to investigate whether integrating subject-specific
brain connectome into regional amyloid-´ pathology (RAP; standardized uptake value ratio in PET) measures
improves predictive validity for cognitive decline.
Methods: Diffusion and resting-state functional MRI were acquired from eighty-five cognitively normal older
adults (age 65.56±3.35). We first obtained subject-specific structural connectivity (SC) and functional connectivity
(FC) (Fig.1). Then, we developed two network-based amyloid-´ pathology (NAP) measures: 1) connectivityweighted NAP scores quantified both RAP and amyloid-´ in all other connected regions, weighted by its
connection strength; 2) centrality-scaled NAP scores quantified the RAP in the region scaled by its centrality in
the connectivity matrix. Using a cross-validated predictive model (Fig.2), we evaluated the performance of NAP
measures in predicting longitudinal cognitive decline in global cognition (4.51±0.68 years follow-up). We compared
these results with those from RAP measures that did not include connectome information, and a benchmark
model with only connectome measures.
Results: Baseline RAP significantly predicted subsequent cognitive decline (permutation p
