Automatic Blood Clot Detection for Extracorporeal Life Support Systems (ECMO Machines)

<p dir="ltr">Extracorporeal Membrane Oxygenation (ECMO) is a life support technique used to treat critically ill patients with cardiac and pulmonary failure. During ECMO, blood is taken from the patient and pumped under pressure to the oxygenator, where it is oxygenated and warmed, and finally returned to the patient. Due to the turbulent flow and contact with artificial surfaces, ECMO activates hemostasis, causing both patient-related and circuit-related clotting. Circuit clotting requiring component exchange is one of ECMO's most frequent and expensive complications. Clot formation in the circuit can result in life-threatening thromboembolisms if they detach and enter the patient. The obstruction of a blood vessel in this manner can lead to cardiac arrest and stroke. </p><p dir="ltr">Hospitals have multiple techniques for monitoring patient and circuit thrombosis. However, our literature review found that these techniques cannot accurately measure the amount of clot formation in a circuit component. Additionally, these techniques cannot predict clot location, which we posit is vital information to reduce the number of complications in ECMO. The best time to exchange ECMO components is also unknown, with protocols varying between hospitals. Improving ECMO clot detection would provide clinicians with more information about the condition of the patient and the circuit components. This information could be used in further studies to find the optimal time to exchange ECMO components and improve patient outcomes. We focused our research on the oxygenator since it is the most challenging component to assess and the most frequent site of clot formation. </p><p dir="ltr">To investigate the full scope of the problem, we completed an extensive literature review on the topic, subsequently published as a review paper. This paper explores why oxygenator clot detection is an essential area of ongoing research in ECMO and analyses the benefits and limitations of current hospital techniques. The paper then assesses solutions proposed by other researchers and presents two promising areas for future work. These areas are the in-depth analysis of pressure signals in the ECMO circuit and the use of real-time imaging techniques to provide continuous and accurate clot detection. </p><p dir="ltr">Following the literature review, this thesis explores how we investigated pressure fluctuations and ultrasound imaging for clot detection in the oxygenator. We detail the construction of a mock ECMO loop to perform customizable experiments using polydimethylsiloxane (PDMS) clot phantoms. We used both real ECMO oxygenators and modifiable 3D-printed prototypes. We describe building an ultrasound clot assessment system, including the numerous electrical and software complications we overcame to produce a functional product. We use time series feature extraction and machine learning to predict the location and geometry of clots within our oxygenators, with promising results. </p><p dir="ltr">We then evaluate our ultrasound system on animal blood, demonstrating the ability to differentiate between clotted and non-clotted pig blood. Finally, this thesis will describe our experiments on an actual ECMO circuit with human blood in collaboration with Monash University. We found that our ultrasound system potentially detected oxygenator clotting before any other hospital technique.</p>