Advancing automated blood flow monitoring using ultrasound
Luuk van Knippenberg defended his PhD thesis at the Department of Electrical Engineering on March 3.
Proper blood flow is essential for health, since blood carries oxygen and nutrients to the body’s organs. That’s why monitoring of the blood flow can help doctors detect early signs that something is going wrong, especially during major surgery or in intensive care. Yet, there is a tradeoff between invasiveness and accuracy in current clinical practice. The most accurate and continuous blood flow measurements require sensors placed inside major blood vessels or the heart, which increases the risk of bleeding and infection. Less invasive methods are safer and easier to use but tend to be less reliable and often only provide intermittent information. Ideally, a new monitoring solution would be non-invasive, continuous, and independent of the operator, allowing it to be used in settings ranging from patient’s homes to the operating room. In his PhD research, Luuk van Knippenberg explores how modern ultrasound technology can meet this challenge.
Ultrasound is an attractive technology for estimating blood flow because it is safe, portable, and widely available. Recent advances have enabled devices small enough to be handheld or even wearable. Instead of placing sensors inside the body, ultrasound sends harmless sound waves through the skin. A particularly suitable target is the common carotid artery, a large vessel that carries blood from the heart to the brain. Located close to the skin and easily accessible on either side of the neck, it provides a direct window into central circulation.
A new imaging approach
Traditionally, carotid ultrasound is performed by aligning the probe along the length of the artery. While effective, this approach is very sensitive to motion and depends heavily on the operator. These limitations make continuous monitoring difficult. In his research, Luuk van Knippenberg introduces and validates an alternative approach: tilting and rotating the ultrasound probe to view the artery in cross-section. In this view, the ultrasound image captures the full width of the vessel and the complete pattern of blood flow velocities across it during each heartbeat. The artery appears as an ellipse, and this shape can be used to compute the angle needed to convert ultrasound signals into true blood flow speeds. This approach is more tolerant of movement and less dependent on operator choices, making it better suited for continuous monitoring.
Automatic monitoring system
For continuous monitoring to work in practice, the ultrasound system must also be able to automatically recognize and track the artery. This is challenging because ultrasound images are noisy and vary in quality between machines. While neural networks have shown strong performance in recent years, they typically require large labeled datasets for training, which are scarce in medical imaging. To overcome this limitation, Knippenberg first trained on simulated images where the correct answers are known. The training was then adapted to real ultrasound scans without labels by using prior knowledge: in cross‑section, the carotid artery must resemble an ellipse. This constraint allows the network to self‑correct during training and reliably track the artery frame by frame without human input. All components were then integrated into a fully automated system that detects the artery, measures its shape, and continuously adjusts ultrasound settings in real time to optimize blood flow measurements. Tests using simulations, experimental setups, and healthy volunteers demonstrated that the approach is feasible and promising for clinical use.
Towards real-world application
Finally, Knippenberg studied how natural variations in artery shape, such as curvature or slight deviations from circularity, affect measurement accuracy using realistic 3D artery models. These results provide practical guidelines for clinical application and show that the method is robust under real-world conditions. Together, these developments lay the groundwork for future wearable ultrasound devices capable of continuous blood flow monitoring. Such technology could enable earlier detection of cardiovascular deterioration, reduce the need for invasive procedures, and bring intensive-care-level insight to more patients.
This research was done in collaboration with Philips Research, Catharina Hospital Eindhoven and NWO (project 17878 – BRUM)
Title of PhD dissertation: . Supervisors: Prof. Massimo Mischi, Dr. Ruud van Sloun and Dr. Jens Muehlsteff.