Dr. Van Laere took the first steps toward developing an AI-based solution by joining forces with a bio-informatics research group at the University of Antwerp. The solution’s first few iterations were financed by a grant from the university. A local researcher working on the project became the first colleague to join Innocens BV, a joint spin-off of the University of Antwerp and UZA.

Dr. Van Laere also discussed possible solutions with his close friend Dirk A. Claessens, an IBM executive, consultant and specialist in AI, data and predictive analytics.
 
The duo frequently traded work stories while on their weekly bike rides through town or over a meal at the local bistro. These get-togethers were a welcome reprieve for Dr. Van Laere, whose schedule normally consisted of caring for newborns in crisis and having tough conversations with fearful parents.
 
It was during these retreats that the two realized they had more in common than a love of cycling and great food—they also shared a passion for data. “Data tells a story. When a patient has severe complications, we can see how their physiology is changing in the data. There must be a way to determine where that story is heading, so we can improve the ending,” Dr. Van Laere said. With this spark of inspiration from the doctor, Claessens began scribbling down ideas.
 
“The solution you want to build has to help detect the potential signs that could indicate adverse outcomes, like sepsis, faster, within neonates based on the data you have,” Claessens said during their spirited discussion. The UZA NICU had a decade’s worth of admissions data on premature and low birthweight infants, giving the two men a strong starting point. Dr. Van Laere wanted to incorporate this data into an AI-enabled predictive solution capable of providing insights to healthcare workers. “My main concern is having the ability to see the signs of a possible infection as soon as possible—even at night, even when our unit is busy.”
 
With the breadth of AI solutions and technical expertise from IBM Consulting in Amsterdam, IBM Research® in Almaden and the IBM Watson® Center in Munich, Claessens knew IBM could be the ideal technology partner for realizing the doctor’s vision. These brainstorming sessions, along with developments from the university research group, eventually led Dr. Van Laere and his team to launch Innocens BV, a subsidiary created to further develop and validate the Innocens solution.

Innocens, which is short for Improving Neonatal Outcome with a Clinical Early Notification System, is an edge computing technology that trains computers to analyze data streams from patients to find patterns that could indicate late-onset sepsis. According to Dr. Van Laere, a solution like Innocens is built on three pillars: a predictive model, a compelling user interface and a robust architecture.

Predictive model

Customers can train computers using a process called machine learning, a subcategory of AI that uses algorithms to learn from data, draw inferences from patterns within it and help predict outcomes. Those algorithms are constantly correcting and training themselves to be faster and more accurate.

IBM Client Engineering helped Innnocens in developing and testing the federated machine learning model that the Innocens BV solution intends to use. Innocens BV used IBM Watson Studio to train its solution’s machine learning models to detect bloodstream infections in infants at the NICU. IBM Watson Studio, a core service on IBM Cloud Pak® for Data, provides a platform to build, run and manage models at scale.

User interface

The user-friendly interface is intuitive and provides insights to be interpreted by the user. “We took advantage of the explainable AI capabilities built into IBM Cloud Pak for Data, the data platform used for the modeling,” Dr. Van Laere explains. “By helping users better comprehend what the models are telling them and why, we’re building a foundation of trust between caregivers and their instruments—a trust that’s imperative if we want to remain vigilant.”

Claessens expounds on the importance of trust. “The user interface is absolutely critical to strengthening the user’s understanding of the technology. We want to provide technology that gives doctors insights which they can then use to inform their diagnosis. The idea is that the computer elevates human insights, but the doctor ultimately maintains control.”

Robust architecture

A robust architecture that integrates edge computing brings computation and data storage closer to the data source. This is crucial in a healthcare setting where sensitive information is shared during the care process and where time is of the essence. “The devices that will do visualization and forecasting need to be in close proximity to the data source and the people using the data,” Claessens states.

Disparate data sources can compromise security and lead to response latency. “You have the hospital, then the patient room inside the hospital, then the devices inside the patient room. We want to wall off each of those areas to help protect the data and process insights in real time,” says Claessens.

Innocens models run locally within the hospital’s firewalls and can function and evolve without removing sensitive data from the hospital. “The raw data will remain on premises. Federated machine learning does this without moving the data. The parameters will move in the cloud, but the raw data will stay within hospital walls,” says Claessens.

The impact of the Innocens technology is being investigated in clinical trials. Commercial availability could occur in the coming years.

What started as a simple exchange of ideas between friends eventually became a ground-breaking approach to neonatal care.

At Innoncens BV, Dr. Van Laere and his team used IBM technology to create a data and AI environment that allows doctors to study patterns, question results and design individualized value-based care.

The predictive model provides doctors with a continuous, explainable and data-driven basis for their care decisions. Dr. Van Laere continues, “Innocens works alongside us to monitor infants around the clock, seven days a week.” By augmenting the intelligence of bedside healthcare workers, NICU doctors can focus on providing comfort and precision care for their patients.

Ultimately, Dr. Van Laere and Claessens see the Innocens solution’s impact on predicting potential early sepsis and treatment as the beginning of a longer journey toward applying AI to improve newborn care. “We hope the same model-driven approach can be used to detect other complications of prematurity at an earlier stage,” says Dr. Van Laere. IBM Cloud Pak for Data, IBM Watson Studio and IBM Watson Machine Learning are powering and underpinning Innocens BV’s plans to deploy the solution in other NICU hospitals and systems around the world.

Innocens BV (link resides outside of ibm.com) is a research and development startup from the Neonatal Intensive Care Unit of Antwerp University Hospital (UZA). Innocens is an acronym for “Improving Neonatal Outcome with a Clinical Early Notification System,” and the Innocens solution aims to develop a clinical decision support system based on AI technology.

¹ Palmeira, P., Quinello, C., Silveira-Lessa, A. L., Zago, C. A., & Carneiro-Sampaio, M. (2012). IgG placental transfer in healthy and pathological pregnancies. Clinical & developmental immunology, 2012, 985646. https://doi.org/10.1155/2012/985646