Computational fluid dynamics (CFD) simulations have been proposed as a tool for the pre-operative evaluation and planning of cardio- and cerebrovascular diseases, such as brain aneurysms. By studying the changes in blood flow induced by pathological vascular defects and linking them to the biological response, we can attempt to predict the time course of the disease before and after treatment. However, CFD simulations are computationally expensive and clinical practitioners have neither the training required nor access to the supercomputing resources needed to utilise such tools currently.
In this project, we apply deep learning through convolutional neural networks to accelerate the fluid simulations by training the network to mimic the outputs of a traditional CFD solvers. Parametric virtual phantoms will be developed for various vascular diseases and the neutral network will learn dependency of the flow and related biological processes based on the parameters of the phantom geometry. Such models can then be translated into applications that provide real-time predictions of vascular flow and support clinical decision making.