Machine Learning for Angiography-Based Blood Flow Velocity Prediction

Blood flow and wall shear stress impact on arterial walls contribute to thrombus formation and plaque cap rupture in arteries. However, it is challenging to measure in-vivo hemodynamic in arteries directly. Computational Fluid Dynamics (CFD) has become an instrumental technique for assessment of detailed blood flow characteristics. However, it involves complex numerical modeling and simulations. In this work, we propose a framework to estimate hemodynamics in blood vessels solely based on angiographic images by employing machine learning (ML) algorithms. In the experimentally validated CFD modeling, the iodine contrast perfusion into blood was simulated using a dye injection into the water flow in a tube. Using CFD simulations, the ground truth velocity field and projective images of flows with dye injections were obtained. Then the rough velocity field was obtained by using the optical flow method (OFM) on projective images. ML models were trained using the least absolute shrinkage selection operator, multilayer perceptron, convolutional neural networks, and long short-term memory networks with CFD velocity as the ground truth and OFM velocity as the input. The performance of each model was evaluated using a set of 613 images based on mean absolute error and mean squared error, where all models outperformed the error of 3*10^-3 m/s and 5*10^-3 m/s respectively. Validation results show that ML models significantly reduced the error rate from 53.5% to 2.5% on average for the v-velocity estimation compared to CFD simulations. Hence, the ML framework provided an alternative pathway to support a clinical diagnosis from hemodynamics perspectives.