Machine learning to predict cardiovascular events and response to therapy based on echocardiographic-derived functional and morphological characteristics of the heart.
Cardiovascular disease remains a major health problem worldwide, as it is responsible for about 30% of all deaths. When diagnosing the heart, ultrasonic imaging remains the modality of choice not only due to the fact that it is non-invasive, mobile and relatively cheap but also because it can generate images in real-time and at a high rate (e.g. conventionally about 30 images/second can be generated). Although worldwide a lot of research efforts focus on estimating cardiac morphological and functional parameters in an accurate and robust manner, little attention has been given to aid the clinician in further interpreting the obtained measurements. Nevertheless, it is well recognized that these data sets are complex and hard to interpret – even by experts.
Within this project, we will take advantage of state-of-the-art machine learning methodologies in order to develop a tool that can support the physician in interpreting echocardiographic data and therefore guide the decision-making process. More specifically, we will extract information on local cardiac function and shape – after correcting them for confounding factors such as age or gender – and determine their (individual and joint) added prognostic power. As first application domains, we will predict the risk of developing future cardiac disease on the one hand and the response to biventricular pacemaker therapy in heart failure patients on the other.