Is it possible to prevent sepsis (blood poisoning) in premature babies by means of Artificial Intelligence? Six students found out during the Fruit Punch Challenge AI.
A baby born prematurely – before the 37th week of pregnancy – has a higher risk of complications. One of those risks is developing sepsis. It is the body’s inflammatory response to an infection that is so severe that tissues are damaged and organ functions fail. This can have serious consequences or the baby can even die.
Even more accurate and better forecasting
UMC Utrecht uses a machine learning (ML) prediction model that can predict the risk of a baby developing sepsis during a stay in the NICU based on heart rate and oxygen saturation. In recent months, Fruit Punch Artificial Intelligence and UMC Utrecht challenged six students to develop a model that can predict sepsis in premature babies even better. The aim is to provide doctors with an accurate risk analysis so that they can quickly take the right measures to limit health damage or death in premature babies. The students did this under the supervision of neonatologist Daniel Vijlbrief and the Digital Health Data science experts O’Jay Medina and Richard Bartels from UMC Utrecht. On Friday 24 June they presented their solutions. They will use the insights gained from the challenge in the development of a newer, improved version of the machine learning (ML) prediction model.
Result: refinement and renewal of existing forecasting models
The students started working in two groups. Resulting in a refinement of the existing logistic regression model, a new XGBoost model and an elaboration with SHAP (Shapley Additive Explanations – a method to interpret the data arising from, for example, the ML prediction model) for XAI (Explainable Artificial Intelligence). “The gains of the past few months are mainly in the new insights that have been generated, which we can look at to improve our algorithm. Both teams have taken a slightly different path in building and calculating the features used to predict the risk of sepsis. We are not using those features yet. Now it’s our job to see how we can integrate this to potentially make our algorithm even better,” said O’Jay Medina.
Participants in the FruitPunch AI challenge
It tastes like more
This was the first time that an AI issue from the UMC Utrecht was solved through a FruitPunch challenge. That tastes like more. The organization is very satisfied and aims to organize such a challenge once or twice a year in future. Annemarie van ‘t Veen, lead applied data science at UMC Utrecht: “Working with data collected in daily clinical practice was experienced as very valuable by students. And the students in turn have inspired us. We call on everyone with a background in data science to participate next time. This can be our students, but also, for example, data scientists working in other industries such as the banking world. The cross-pollination of knowledge is invaluable to all of us.”
Artificial Intelligence (AI) has the potential to provide solutions to the world’s biggest problems, but it is currently underused. There are not enough people who know how to apply AI. The FruitPunch AI develops challenges for students to teach them about AI, based on their own experience and while developing their own skills. Together with Utrecht University, Eindhoven University of Technology and Wageningen University, UMC Utrecht stimulates the use of AI by supporting and guiding these challenges.
Over Digital Health
UMC Utrecht is strongly committed to healthcare innovation. We do this by creating a seamless care process for patients together with patients, general practitioners and other healthcare providers and by giving patients more control themselves. We use ICT and mobile eHealth technology, such as apps and home monitoring equipment. In this way the patient gets more self-management at home and is supported by care when it is really necessary or desired with more person-oriented advice and better coordination between the care providers.
Look here for the most recent Data Science and eHealth solutions from UMC Utrecht.
Questions, comments or tips for the editors?