Informing 4D flow MRI haemodynamic outputs with data science, mathematical models and scale-resolving computational fluid dynamics

在英国，心脏病导致的死亡约占所有死亡人数的四分之一，相关的医疗成本估计为每年90亿英镑。主动脉是人体内最大的动脉，与心脏直接相连。主动脉疾病是最常见的心血管疾病之一，可能会极大地危及生命。管壁剪应力是血流对动脉壁内表面施加的剪切力，是动脉壁疾病的重要生物标志物。同样，血液中的血液流动障碍，如湍流，也与心脏病有关。4D血流磁共振成像(MRI)是一种测量动脉血流速度的MRI序列，在临床和心血管研究中都有广泛的应用。在临床上，MRI被用来诊断心脏病和评估治疗方法。目前，4D Flow MRI无法准确评估壁面剪应力和湍流等重要指标，但随着新方法的发展，这是可能的。在这项研究中，我将开发新的方法，从根本上提高4D Flow MRI的血流动力学输出能力。这将使用计算流体力学、数据科学和机器学习方法来实现。MRI序列具有由用户确定的各种设置，目前，设置是针对速度场采集而优化的，而不是壁面剪应力或湍流。为了能够准确测量这些参数，需要建立最佳的MRI设置。我将为‘虚拟’4D Flow MRI开发一个工具，它可以复制真实的4D Flow MRI序列。然后，该工具将被用于虚拟地优化4D Flow MRI序列。接下来，我将开发一个基于超分辨率的机器学习模型，该模型可以改善4D Flow MRI血流动力学输出。该模型将使用各种动脉的高分辨率和低分辨率计算流体动力学模拟的组合进行训练。开发的模型将使用新的多分辨率4D血流MRI扫描和高分辨率计算流体动力学主动脉模拟进行验证。这项研究将通过与曼彻斯特大学机械、航空航天和土木工程系和心血管科学系的专家和最先进的设施合作进行开发。这个多学科项目将吸收英国研究人员、临床医生和生命科学家的专业知识，并利用BHF曼彻斯特心肺磁共振研究中心的磁共振设施，实现新的扫描采集。总的来说，这项研究将推动数据驱动的流体动力学技术和磁共振成像方法的发展，直接影响医疗保健部门，并使流体动力学、数据科学和心血管研究社区受益。在研究中，改进的4D Flow MRI将使更大规模的研究成为可能，并可能改善利用MRI数据进行的计算流体动力学研究。在临床环境中，从MRI获得更广泛的血流动力学参数将有助于开发新的诊断工具和新的治疗工具，改善患者的预后。在项目结束时，我将开发新的方法来改进不同阶段的4D Flow MRI输出-从MRI采集到图像后处理。

In the UK, heart diseases cause around a quarter of all deaths and related healthcare costs are estimated at £9 billion annually. The aorta is the largest artery in the body, connected directly to the heart. Aortic disease is one of the most common types of cardiovascular disease and can be extremely life threatening. Wall shear stress is the shearing force exerted by blood flow on the inner surface of the arterial wall and is an important biomarker for arterial wall diseases. Similarly, blood flow disturbances in the blood like turbulence are also linked with heart disease. 4D flow magnetic resonance imaging (MRI) is a type of MRI sequence which measures blood flow velocities in arteries and is used both in clinic and in cardiovascular research. In clinic, MRI is used to diagnose heart disease and evaluate treatments. Currently, important metrics like wall shear stress and turbulence cannot be accurately evaluated from 4D flow MRI, although it may be possible with the development of new methods.In this research fellowship I will develop new methods to radically improve the haemodynamic output capabilities of 4D flow MRI. This will be achieved using computational fluid dynamics, data science and machine learning methods. MRI sequences have various settings determined by the user and currently, settings are optimised for velocity field acquisition, not wall shear stress or turbulence. To enable accurate measurement these parameters, optimal MRI settings need to be established. I will develop a tool for 'virtual' 4D flow MRI which can replicate real 4D flow MRI sequences. The tool will then be used to optimise 4D flow MRI sequences virtually. Next, I will develop a super resolution-based machine learning model which can improve 4D flow MRI haemodynamic outputs. The model will be trained using a combination of high-resolution and low-resolution computational fluid dynamic simulations of various aortas. Developed models will be validated using new multi-resolution 4D flow MRI scans and high-resolution computational fluid dynamic aorta simulations.This research will be developed via collaboration with experts and state-of-the-art facilities found within the Department for Mechanical, Aerospace and Civil Engineering and the Division of Cardiovascular Sciences at the University of Manchester. The multidisciplinary project will draw expertise from researchers, clinicians and life scientists in the UK, as well as make use of MRI facilities at the BHF Manchester Centre for Heart & Lung Magnetic Resonance Research enabling new scan acquisitions. Broadly, this research will advance data-driven fluid dynamics techniques and MR imaging methods, directly impacting healthcare sectors as well as benefitting fluid dynamics, data-science and cardiovascular research communities. In research, improved 4D flow MRI would enable larger-scale studies and likely improve computational fluid dynamic studies informed with MRI data. In clinical settings, access to a wider range of haemodynamic parameters from MRI would enable development of new diagnostic tools and new treatment tools for aortic disease, improving patient outcomes. At the end of the project, I will have developed novel methods to improve 4D flow MRI outputs at different stages - from MRI acquisition through to image post-processing.