Understanding Blood Flow Dynamics and its Impact on Health

Changes in blood flow can significantly impact the structure, function, and overall health of arterial walls and the heart. These changes can occur because of a heart attack or they can contribute to the development of life-threatening aneurysms and other harmful conditions. For clinicians, understanding how their patient’s unique blood flow dynamics might contribute to disease and using this information to determine the right course of treatment, such as medication or surgery, is often limited by the accuracy of clinical imaging techniques and the computational power available.  

To address this unmet need, researchers at the Leeds Institute for Fluid Dynamics (LIFD) at the University of Leeds are harnessing advancements in machine learning (ML) and artificial intelligence (AI) alongside the latest imaging technologies. By combining these innovations with a strength of multidisciplinary academic expertise across the Institute, they are working to enhance the accuracy of imaging techniques and the effectiveness of computational models, so swift and reliable calculations can be made about blood flow and its effects on the human body.  

Predicting cardiac remodelling after a heart attack 

Following a heart attack, the heart can undergo the process of cardiac remodeling, where the left ventricle can gradually shift in size and shape over a period of several months. This change in heart geometry can cause the heart to be less efficient and is a precursor of serious heart disease. Typically, assessing for long-term damage involves scanning a patient’s heart within hours of hospital admission, with follow-up scans needed for up to two years. Now, Leeds-led research is working to understand the risk of further complications for recovering heart attack patients from just their initial heart scan.  

A study led by Dr Erica Dall’Armelina in the School of Medicine, with researchers across the LIFD, is delivering new insights into blood flow beyond what can be achieved from imaging alone. The team are applying 4D Flow Magnetic Resonance Imaging (MRI) and ML to accurately calculate the pressure and velocity of blood flow through the left ventricle of the heart. Typically, MRI scans are limited by low resolution. The researchers are applying a Physics Informed Neural Network technique, initially developed by Leeds computer scientists, to provide clearer images of the flow in the left ventricle so more data can be obtained. This is then automatically post-processed to rapidly calculate stresses on the heart wall. 

The aim is to use these detailed blood flow models to assess whether early signs of high shear stress on the heart wall can predict the risk of cardiac remodeling. By identifying those at higher risk of complications, this approach could help guide preventative treatment for heart attack patients. 

Accelerating aneurysm treatment decisions 

Aneurysms in the brain represent another critical challenge for clinicians, forming when weakened blood vessel walls bulge. Many people may live unaware of their aneurysm, but if it ruptures, an aneurysm can have life-threatening consequences or lead to permanent brain damage. For each patient, clinicians need to determine the risk of rupture, and will evaluate the different treatment options, like fitting a stent or managing immediate risk factors such as high blood pressure. This involves scanning the shape and geometry of an aneurysm, calculating its effect on blood flow and predicting its behaviour. These simulations, based on computational fluid dynamics and traditional mathematical models, can take many hours to run, but for many patients, time is critical.  

To address this, fluid dynamics researchers at Leeds, in collaboration with Professor Alejandro Frangi’s team at the University of Manchester, are incorporating ML to significantly speed up these 3D simulations. The team are using ML-augmented Reduced Order Models to improve simulation performance, designed to reduce the complexity of large, computationally expensive and time-consuming calculations. While this is early-stage research, the goal is to produce accurate and real-time results for an individual patient in seconds, instead of hours.  

For this project, overcoming computational complexity and the need for real-time decision support for clinicians for each individual patient requires a network of clinicians, computer scientists, mathematicians, mechanical engineers and biomedical engineers. To be clinically useful, the models need to be tailored to the specific geometry and anatomy of each patient’s aneurysm, and the team are also faced with the complex challenges of transforming medical images into precise computer models for blood flow simulations.  

Peter Jimack, Professor of Scientific Computing at the University of Leeds, said “Machine learning is giving us an extra tool in the box for solving classical problems. Where we would typically use numerically-based simulation techniques to run calculations based upon a model of the physics alone, we can now replace or enhance these with data-driven machine learning models to create much faster and more accurate representations and predictions of blood flows from medical images. 

“With advancements in hardware and innovative machine learning methods, we're able to move beyond traditional computational mechanics. This progress opens up possibilities that weren’t feasible a decade ago, with the potential to improve medical imaging, accelerate surgical simulations, and increase the success rates of clinical trials for new medical devices.”