Neural Operators for Detecting Aortic Aneurysm Contributors
DescriptionThoracic aortic aneurysm is a localized dilatation of the aorta that can lead to life-threatening dissection or rupture. Currently, the only way to assess the progression of this condition is by measuring the size of the aneurysm and its growth rate. However, there is hope that by using computational modeling of the aorta's biomechanics, we can predict its future geometry and properties based on the initial causes of the condition. To achieve this, we have created a framework that uses a deep operator network-based surrogate model to identify factors that contribute to thoracic aortic aneurysms using synthetic finite-element-based datasets. We have trained our model by generating maps of local aortic dilation and distensibility for multiple risk factors using a constrained mixture model of aortic growth and remodeling. The performance of two proposed frameworks is evaluated on insult distributions ranging from simple to complex. We have found that our continuous learning approach can accurately predict a patient's specific insult profile associated with any given dilation and distensibility map, especially when based on greyscale full-field images. Our results suggest that DeepONet can be used to support transfer learning of patient-specific inputs to predict the progression of thoracic aortic aneurysms.
TimeTuesday, June 2711:00 - 11:30 CEST
LocationSanada I
Event Type
Life Sciences