CAREER: Synergistic physics-based and deep learning cardiovascular flow modeling

职业：基于协同物理和深度学习的心血管血流建模

• 批准号: 2247173
• 负责人: Amirhossein Arzani
• 金额: $ 50.76万
• 依托单位: University of Utah
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2027-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2247173&HistoricalAwards=false
• 关键词: CAREER; Synergistic; physics; based; deep

Accurate quantification of blood flow across different scales is crucial to our fundamental understanding of cardiovascular disease and clinical decision making. While computational and experimental blood flow modeling has seen tremendous progress, we still have difficulty generating reliable data. Low resolutions and unknown parameters overburden high fidelity modeling. Additionally, blood flow dynamics near the wall where disease localizes are hard to quantify due to thin boundary layers and challenges in near-wall transport modeling. Finally, the large datasets that blood flow models produce are difficult to efficiently store and interpret. This project will develop software that synergistically integrates novel deep learning and traditional physics-based modeling to address these issues. The software will promote the progress of scientific cardiovascular disease and fluid flow modeling research and ultimately advance national health. The project will create new education programs integrated with research to promote fluid flow computer modeling education by blending data analysis, visualization, and computer modeling. The education program will be integrated with regional initiatives to promote STEM participation in underrepresented groups.This CAREER program is built on four overarching goals. First, physics informed neural network (PINN) models will be used to overcome inherent blood flow modeling limitations. Second, the models will be used for gaining a physical understanding of blood flow patterns across different scales. Auxiliary PINN models will be defined to bridge blood flow patterns away and near the vessel wall and understand the minimal data collection necessary for near-wall blood flow modeling. Subsequently, a hybrid computational fluid dynamics (CFD) and PINN model will be developed for on-the-fly CFD and deep learning modeling. The goal is to compress and store the wealth of information that is generated and often ignored by CFD solvers and tackle difficult multiscale problems that challenge traditional CFD approaches. Finally, an education program called FAST (Fluids, Art, and StoryTelling!) will be developed to generate enthusiasm for integrated computer modeling and engineering education. The goal is to leverage the art of visualization and storytelling to show the hidden beauty in fluid mechanics computer modeling. This CAREER program will build a foundation for hybrid and complementary deep learning and physics-based modeling approaches that advance fundamental and transformative blood flow modeling research and will enable lifetime leadership in integrating research with education.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

准确量化不同尺度的血流量对我们对心血管疾病和临床决策的基本理解至关重要。虽然计算和实验血液流动模型已经取得了巨大的进步，但我们仍然难以生成可靠的数据。低分辨率和未知参数影响了高保真度建模。此外，由于薄边界层和近壁输运建模的挑战，疾病所在的壁面附近的血流动力学很难量化。最后，血流模型产生的大型数据集很难有效地存储和解释。该项目将开发软件，将新型深度学习和传统的基于物理的建模协同集成，以解决这些问题。该软件将促进心血管疾病和流体流动模型研究的科学进展，最终促进国民健康。该项目将创建新的教育项目，结合研究，通过混合数据分析、可视化和计算机建模来促进流体流动计算机建模教育。该教育计划将与区域倡议相结合，以促进代表性不足的群体参与STEM。这个职业计划建立在四个总体目标之上。首先，物理信息神经网络（PINN）模型将用于克服固有的血流建模限制。其次，这些模型将用于获得对不同尺度的血液流动模式的物理理解。将定义辅助的PINN模型，以桥接血管壁外和血管壁附近的血流模式，并了解近壁血流建模所需的最小数据收集。随后，将开发一种混合计算流体动力学（CFD）和PINN模型，用于实时CFD和深度学习建模。目标是压缩和存储CFD求解器经常忽略的生成的丰富信息，并解决挑战传统CFD方法的困难多尺度问题。最后，一项名为FAST（流体、艺术和讲故事！）的教育计划将被开发出来，以激发人们对综合计算机建模和工程教育的热情。我们的目标是利用可视化和讲故事的艺术来展示流体力学计算机建模中隐藏的美。该职业计划将为混合和互补的深度学习和基于物理的建模方法奠定基础，推进基础和变革性的血流建模研究，并将使研究与教育相结合的终身领导成为可能。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Ensemble physics informed neural networks: A framework to improve inverse transport modeling in heterogeneous domains

集合物理通知神经网络：改进异构域逆向传输建模的框架

• DOI: 10.1063/5.0150016
• 发表时间: 2023
• 期刊: Physics of Fluids
• 影响因子: 4.6
• 作者: Aliakbari, Maryam;Soltany Sadrabadi, Mohammadreza;Vadasz, Peter;Arzani, Amirhossein
• 通讯作者: Arzani, Amirhossein

Theory-guided physics-informed neural networks for boundary layer problems with singular perturbation

用于解决奇异扰动边界层问题的理论指导物理信息神经网络

• DOI: 10.1016/j.jcp.2022.111768
• 发表时间: 2023
• 期刊: Journal of Computational Physics
• 影响因子: 4.1
• 作者: Arzani, Amirhossein;Cassel, Kevin W.;D'Souza, Roshan M.
• 通讯作者: D'Souza, Roshan M.