Speaker: Jacob Vogel (VUmc Alzheimer Centre, Amsterdam)
Title: A data-driven approach to summarizing tau-PET signal covariance in Alzheimer's disease
The use of positron emission tomography (PET) to target filamentous tau proteins represents a fairly new approach to in vivo imaging of pathological proteins in the brain of people with Alzheimer’s disease. Researchers continue to search for measures that summarize the information contained in tau-PET images in ways that maximizes effects while minimizing dimensionality. Previous attempts at finding such measures have been mostly based on autopsy studies measuring the spatial extent of pathological tau fibrils in the cerebral cortex. However, these ex vivo approaches present limitations when applied to in vivo tau-PET images. In such cases, exploratory data-driven approaches that allow the data to “describe itself” may be informative, both about the true distribution of tau-PET signal, but also regarding the optimization of tau-PET biomarkers for predicting clinical outcomes. We use an unsupervised clustering algorithm called Bootstrap Analysis of Stable Clusters (BASC) to identify independent sources of cross-subject covariance in tau-PET images. We compare these data-driven clusters to autopsy-derived regions described in previous literature, and we show enhanced performance of the data-driven clusters in predicting cognition in a separate sample. The results suggest data-driven approaches may be informative in ongoing investigations of pathological tau and its impact in Alzheimer’s disease. In addition to presenting our study, I will also briefly describe some of the open-source tools used for this analysis.