A computational approach combining 4D Flow and CFD for improved determination of cerebral hemodynamics

项目摘要 该项目旨在生成超分辨率时间分辨相衬MRI（4D flow MRI），以改善 脑血流动力学的定量。我们的团队已经在其他几个实验中收集了大量的体内4D Flow数据。 资助的项目。这些数据是用传统的手工方法处理的， 适度的收购决议。我们有三种复杂程度的数据。首先是极高质量的数据 在使用患者特定几何形状和流动条件创建的非常仔细控制的流动模型中。二是 颅内静脉流出道的体内数据，收集自58例脉动性耳鸣（PT）患者。在 在这些区域，流动具有相对较小的脉动性，但在几何上复杂，具有明显的涡度。 最后，我们获得了148例颅内动脉瘤（IA）患者的4D血流数据。使用这些现有的数据集， 我们提出了一个渐进的进展，以发展先进的方法，提高4D流分辨率。 有令人信服的证据表明，血流动力学参数在确定 血管疾病进展的演变和对治疗的反应。原则上，4D Flow MRI可用于 通过心动周期确定三维速度场。然而，使用可接受的 采集时间，分辨率是不够的，可靠的速度测绘给定的小口径的 颅内血管患者特异性计算和实验模型可以提供上级分辨率，但 它们的准确性取决于模型简化和假设。我们建议解决目前 通过开发一种基于深度学习的超分辨率方法来克服4D血流图像的局限性。在这种方法中， 将使用4D Flow MRI对脑血管中的血流进行成像，并使用患者特定计算 流体动力学（CFD）。 在这项研究中，我们将首先在患者特定数据中生成血流动力学的3D CFD模拟。那就 使用CFD数据开发超分辨率神经网络，以提供更高分辨率的4D流数据。成功 这一项目的完成将提供一个评价工具，经验证，可改善 脑血流动力学像这样的工具可以用来分期介入治疗， 结果，直接支持国家心脏，肺和血液研究所的使命。

PROJECT SUMMARY This project aims to generate super-resolution time-resolved phase-contrast MRI (4D flow MRI) for improved quantification of cerebral hemodynamics. Our team has collected extensive in vivo 4D Flow data in several other funded projects. That data was processed with conventional manual methods and was limited by the relatively modest acquisition resolution. We have data of three levels of complexity. The first is extremely high-quality data in very carefully controlled flow models created with patient-specific geometries and flow conditions. Second, is in vivo data from the intracranial venous outflow tract, collected from 58 patients with pulsatile tinnitus (PT). In these territories, the flow has relatively little pulsatility but is geometrically complex with pronounced vorticity. Finally, we have 4D Flow data from 148 patients with intracranial aneurysms (IA). Using these existing datasets, we propose an incremental progression to develop advanced methods for improving 4D flow resolution. There is compelling evidence that hemodynamic parameters are of major importance in determining the evolution of vascular disease progression, and response to therapy. In principle, 4D Flow MRI can be used to determine the velocity field in three-dimensions and through the cardiac cycle. However, using acceptable acquisition times, the resolution is insufficient for reliable velocity mapping given the small caliber of the intracranial vessels. Patient-specific computational and experimental models can provide superior resolution, but their accuracy depends on modeling simplifications and assumptions. We propose to address the current limitations of 4D flow images by developing a deep learning based, super-resolution approach. In this approach, the flow in cerebral vessels will be imaged with 4D Flow MRI and simulated with patient-specific Computational Fluid Dynamics (CFD). In this study, we will first generate 3D CFD simulation of hemodynamics in patient-specific data. Then, we will develop a super-resolution neural network using CFD data to provide higher resolution 4D flow data. Successful accomplishment of this project will provide an evaluation tool that is validated for improved quantification of cerebral hemodynamics. A tool such as this could be used to stage interventional treatments and improve patient outcomes, in direct support of the National Heart, Lung, and Blood Institute mission.