Shape-driven deep neural networks for fast acquisition of aortic 3D pressure and velocity flow fields

Abstract

Computational fluid dynamics (CFD) can be used to simulate vascular haemodynamics and analyse potential treatment options. CFD has shown to be beneficial in improving patient outcomes. However, the implementation of CFD for routine clinical use is yet to be realised. Barriers for CFD include high computational resources, specialist experience needed for designing simulation set-ups, and long processing times. The aim of this study was to explore the use of machine learning (ML) to replicate conventional aortic CFD with automatic and fast regression models. Data used to train/test the model consisted of 3,000 CFD simulations performed on synthetically generated 3D aortic shapes. These subjects were generated from a statistical shape model (SSM) built on real patient-specific aortas (N = 67). Inference performed on 200 test shapes resulted in average errors of 6.01% ±3.12 SD and 3.99% ±0.93 SD for pressure and velocity, respectively. Our ML-based models performed CFD in ∼0.075 seconds (4,000x faster than the solver). This proof-of-concept study shows that results from conventional vascular CFD can be reproduced using ML at a much faster rate, in an automatic process, and with reasonable accuracy.

A physics-based machine learning technique rapidly reconstructs the wall-shear stress and pressure fields in coronary arteries

“…In the current coronary flow simulations, we have considered mesh shape variability as an evolutionary factor for each steady solution component of interest. This is similar to the recent application of Principle Component Analysis (PCA) to data-driven modelling of aortic flows (23), where separate DNN models were used for pressure and absolute velocity. However, in comparison to the standard PCA and DNN techniques, the suggested cPOD approach allows for extension of the solution matrix from single scalars to 3D velocity vectors and pressure components simultaneously on different meshes in space and time.…”

“…A better balance between number of real data versus synthetic data is required to bring this technique closer to realworld application. In future, a systematic procedure can be adapted to generate the synthetic meshes in an optimal way by exploiting sensitivity of the coronary flow response to perturbations of the baseline vessel geometry, similar to the deformation matrix method recently developed for aortic flow simulations (23).…”

Towards Reduced Order Models via Robust Proper Orthogonal Decomposition to capture personalised aortic haemodynamics

“…This would allow, for instance, the study of a wide range of parameters for a given vascular pathology (e.g. increasing or decreasing the level of stenosis on coronary disease or coarctations) and to analyse the consequences on the flow and pressure fields, which could serve as an initial step to investigate patient-specific pre-interventional options ( Siena et al, 2023 ; Pajaziti et al, 2023 ; Liang et al, 2020 ).…”

Aneurysmal growth in type-B aortic dissection: assessing the impact of patient-specific inlet conditions on key haemodynamic indices

“…By reconstructing WSS distributions accurately in a reduced-order format, our work suggests that POD analysis may also offer opportunities for 4DMR data enhancement and rapid aortic flow reconstruction, providing high-fidelity haemodynamic data within clinically relevant timescales [ 20 , 41 ]. Furthermore, if changes in spatial and temporal mode behaviour with increased inlet flow rate are consistent and predictable across a wider patient cohort, POD analysis may be used to perform efficient uncertainty quantification, perhaps in combination with machine learning techniques.…”