Fast Surrogate Solvers for Flow Networks

Computational models of the cardiovascular system can help gain a better insight into human physiology. We have developed solvers that can provide pressure and flow waveforms across all the large vessels of the vascular system and in the main peripheral beds. However, these models inherently come with a number of parameters each with their own uncertainty. The more complex these models, the more of these parameters need to be estimated and the larger the computing time. Furthermore, when personalising these models there are inherent uncertainties in the measurements. This project seeks to build extremely fast surrogate solvers using deep learning approaches that can provide us with instant diagnostic modelling predictions based on clinical measurements, whilst also providing confidence levels to those predictions. This work is funded by UK Atomic Energy Authority (UKAEA) and the work is conducted in our group by Alex Drysdale.