Researcher at ewuu are developing machine learning models that could help with personalized prognosis for preterm infants. This way, they can start early interventions to reduce the risk of a neurodevelopmental disability later in life. Safeguarding the health of babies with a difficult start.

At UMC Utrecht’s neonatal intensive care unit, up to 60-80 babies are born each year before the 28th week of pregnancy. These extremely preterm infants face a high risk of brain damage and long-term developmental challenges, such as cognitive, motor and behavioural impairment. Yet predicting which baby will face which outcome remains a complex task.

“We often don’t know which baby will develop problems later on,” says Dr. Maria Luisa Tataranno, neonatologist and clinical researcher at UMC Utrecht. “That’s difficult for families, and also for us as doctors. It makes it harder to decide which children need intervention and intensive follow-up and which families we can reassure.”

“We want to give each child an individual prognosis; not just statistics, but real guidance for families and professionals“

-Dr. Maria Luisa Tataranno, neonatologist and clinical researcher at UMC Utrecht.

Small window of opportunity

The preterm brain is very vulnerable, because major transformations happen in the last trimester of gestation. “The preterm brain is still very plastic,” Tataranno explains. “This also means there is a window of opportunity in those early weeks and months when early interventions can make a big difference in  reducing the risk of  neurodevelopmental disability later in life.  “But nowadays, we often don’t know something is wrong until symptoms appear, sometimes years later. That’s too late.” Therefore, there is a need for an early personalized prognosis for the individual patient.

Personalized care

To change that, the PROMISE team is building a model that uses clinical data, EEG recordings(measuring brain activity), MRI scans and even nutritional records, all collected in the first three months of life, to predict outcomes by the age of two. These insights could help clinicians tailor care more precisely, minimize unnecessary procedures, and launch timely therapies that might prevent more serious impairments.

Outcome prediction

The PROMISE project brings together medical experts, AI researchers and data scientists from across the EWUU Alliance (TU/e, WUR, UU and UMC Utrecht) to improve prognosis for vulnerable newborns. Their goal? To create a machine learning model that can predict long-term developmental outcomes based on early-life medical data. While at the same time ensuring that this model conforms to the state-of-the-art values of trustworthy AI.

Machine learning

The team started by creating a FAIR-compliant database (Findable, Accessible, Interoperable, Reusable), harmonizing data collection across departments. “Before, data were collected on paper and passed between teams,” says Tataranno. “Now, we’ve automated much of it. We have EEG scores, MRI injury assessments, and brain volume measurements for around 600 infants — a solid foundation for machine learning.”

First results are promising

The first results are promising. A previously study published in Lancet Digital Health showed that EEG signals from just the first three days after birth could already help predict developmental outcome. Preliminary findings now indicate that combining this data with MRI volumes improves the model’s accuracy. “We’re finalizing the analysis now,” she says, “but we’ve clearly seen that the two modalities are complementary in predicting the outcome.”

Worldwide service

Ultimately, the team hopes to create an open-access web-based service that clinicians worldwide can use to support decision-making. An AI could not replace a doctor, but if we can offer a model where they upload an MRI and get meaningful insights, that could really help.”

For Tataranno, the project wouldn’t be possible without the EWUU collaboration. “I’m a simple doctor,” she laughs. “I can explain why this matters, but I couldn’t build the algorithms. The engineers and data scientists — they make the magic happen.”
And her dream? “To give each child an individual prognosis; not just statistics, but real guidance for families and professionals,” she says. “That’s what we’re working towards.”

About the project

Project title
PROMISE: PRediction of Outcome with Machine Learning in Infants: a Synergistic Exploration

Subject
This research project aims to develop a prediction model for preterm born infants at high risk of behavioural impairment. And to ensure that this is conform to the state-of-the-art values of trustworthy AI. Once validated, this prediction model will be made accessible to a larger network of clinicians through a web-based service.

Research team
Mykola Pechenizkiy (TU/e), Rick Bezemer (TU/e), Clara Belzer (WUR), Chantal Kemner (UU), Albert Salah (UU), Maria Luisa Tataranno (UMC Utrecht), Manon Benders (UMC Utrecht)