Machine Learning Techniques for Predicting Blood Flow and Cancer Cell Trafficking in Microcirculation

血液和组织之间的氧气交换主要发生在微循环中，这是由狭窄的毛细血管组成的。微循环还与许多疾病有关，如心脏和大脑疾病、糖尿病和癌症。红细胞充当氧气载体。它们在微循环中的分布对健康功能和疾病进展至关重要。在过去的几十年里，预测毛细血管中红细胞运动的复杂计算机模型已经出现。然而，这种模型通常需要大量的计算时间和资源，并且对于许多心脏周期的器官尺度预测是不可行的。为了克服这些限制，本提案将考虑应用人工智能或机器学习（ML）技术来预测血流和细胞运输。更广泛的影响将通过拟议的研究，研究生和本科生的培训，以及K-12外展来实现。具体的研究目标包括(i) ML模型的发展，以预测血液流速和红细胞浓度的血管横截面变化，（ii） ML模型耦合血流量，细胞体积分数和细胞变形的单参数测量，（iii） ML模型预测全3D变形细胞形状，（iv）高精度长时间序列预测模型。模型将使用不同几何形状和血流动力学参数的细胞分流和血管的高保真模拟数据进行训练，创建一个模型库。训练后的ML模型将在生理上真实的血管几何中对红细胞流动的高保真模拟进行验证。最后，将通过考虑高保真模拟不可行的大规模脑血管来证明ML模型的能力。该提案的翻译方面涉及开发用于预测癌细胞贩运的类似ML模型。提出的ML模型可以显著提高患者特异性（精度）诊断。博士和本科生将接受指导。外联部分将包括为初高中女生编码和计算。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。

The exchange of oxygen between blood and tissue primarily occurs in microcirculation, which is made of narrow capillary vessels. Microcirculation is also associated with many diseases, such as cardiac and cerebral disorders, diabetes, and cancer. Red blood cells act as oxygen carriers. Their distribution in the microcirculation is critical to healthy functioning and disease progression. Sophisticated computer models to predict red cell motion in capillary vessels have emerged over the past decades. However, such models often require extensive computing time and resources and are not feasible for organ-scale prediction over many cardiac cycles. To overcome such limitations, this proposal will consider the application of Artificial Intelligence or Machine Learning (ML) techniques for the prediction of blood flow and cell trafficking. The broader impact will be accomplished through the proposed research, training of graduate and undergraduate students, and K-12 outreach.Specific research objectives include the developments of (i) ML models to predict vessel cross-sectional variation of the blood velocity, and red cell concentration, (ii) ML models coupling the blood flow, cell volume fraction and a single-parameter measure of cell deformation, (iii) ML models to predict full 3D deformed cell shape, and (iv) models for high-accuracy long time-series prediction. Models will be trained using high-fidelity simulation data of cell flow in bifurcations and vessels of varying geometry and hemodynamic parameters, creating a model bank. The trained ML models will then be validated against high-fidelity simulations of red cell flow in physiologically realistic vascular geometry. Finally, the capability of the ML models will be demonstrated by considering large-scale cerebral vasculatures for which high-fidelity simulation is not feasible. The translational aspect of the proposal involves the development of similar ML models for predicting cancer cell trafficking. The proposed ML models may provide a significant improvement in patient-specific (precision) diagnosis. PhD and undergraduate students will be mentored. The outreach component will involve coding and computing for middle and high school female students.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.