Data-Model Integration in Personalized Computational Medicine
Please login to view abstract download link
A persistent challenge in personalized computational medicine and digital twinning is the integration of sparse, irregular, and multimodal clinical data into physics-based or data-driven models. In this talk, we discuss two methodological approaches we have developed to address this challenge in application scenarios related to brain and cardiac function. First, we introduce a neural network-based disease progression model learned from sparse, multimodal data. Applied to Alzheimer’s disease, this model predicts an individual patient’s future disease trajectory from data collected up to the most recent visit and is able to update its predictions as new data become available. Second, we show how patient risk can be stratified using physics-based numerical simulations. Specifically, cardiac electrophysiology models, personalized using imaging data, are used to numerically quantify the arrhythmic propensity in patients at risk of acute myocardial ischemia [2] or with non-ischemic fibrosis [3]. Together, these applications highlight the translational potential of mathematical and computational models in personalized medicine.
