AI Scans of Preemie Baby Brains Could Block Future Disease

At the BABA Center in Finland, researchers routinely hook up the scalps of newborns with an array of electrodes so they might get an idea of what’s going in inside their brains. The goal is simple but profound: to better understand early neurological development in order to advance better treatments while babies are still young that will ensure more healthy babies.

Late last year, researchers at the center announced a breakthrough published in Scientific Reports that they had built artificial intelligence software that can interpret electroencephalography (EEG) readings from a preterm infant and tell whether the brain is functioning as expected at specific intervals.

This is big, since doctors are limited in the ways they can test what’s going on in a very young brain. It’s not like they can ask the baby what year it is or to touch their nose and expect a response. That why tools like EEG are so important. While an MRI can tell you what the brain physically looks like, an EEG can tell doctors what’s actually going on.

“Clinicians are really more interested in function that structure, because it’s function that they want to preserve,” Nathan Stevenson, a researcher with BABA and the University of Helsinki, which led the software development, told The Daily Beast. “They don’t care if your brain is growing out of your elbow—as long as it’s working perfectly, then that’s all that matters.”

But EEGs aren’t perfect. For starters, there are the technical difficulties in getting numerous electrodes securely in contact with a squirmy newborn head for enough time to get an appropriate reading, which can take hours—even days—depending on the application. This process has been helped significantly by special electrode caps, now available in sizes down to the tiniest preemie. A technician secures the hat by its strap, and squirts a gel between each electrode and the skin, so that the electrodes can detect tiny bits of voltage sent out by bundles of neurons firing in the brain.

But brains are complicated organs, and even more so in little bodies. Because of that, even a high-quality EEG reading is a bit of a mess. “The EEG is a rubbish signal to interpret—it’s just wavy lines,” Stevenson said. Indeed, a readout looks like something that came off a polygraph or seismograph, but with more variables. “You do need some expertise to determine what’s going on.”

Premature babies might stand to benefit most

These days, EEGs are mainly used in neonatal intensive care units as a tool to diagnose seizures. But Stevenson and his colleagues foresee a day when they do much more than that, functioning like a growth chart for the brain and informing doctors in real time about the progress of a baby’s brain growth.

The new software works by reading an EEG recording and comparing it to other, past readings of other baby brains. Brain patterns evolve quickly in newborns, and the computer can tell how mature the signals are. Brain injuries tend to make EEG readings look immature, so if the software output says the baby has a maturational age of 28 weeks, but the infant’s gestational age is 36 weeks, it could indicate that something has gone seriously wrong.

“Pre-term babies are in the intensive care unit for a very long time, and if you can get them to term-equivalent without any incidents, they tend to go on and have perfectly normal lives,” Stevenson said. “Monitoring tools that can help that happen are really useful things.”

Premature birth raises the risk of many complications, including hemorrhages, serious infections, and stroke. Although it’s not yet proven, EEG monitoring could in some cases provide early warning that something has gone wrong in brain development, allowing doctors to intervene faster.

But the potential of EEGs goes well beyond helping with treatment decisions for an individual baby. By offering real-time monitoring of brain development, EEGs could transform research and development and lead to entirely new interventions to ensure the healthy development of preemies.

“It would apply to every single intervention trial that we run nowadays,” Sampsa Vanhatalo, director of the BABA Center and a professor of clinical neurophysiology at Helsinki University Hospital’s Children’s Hospital, told The Daily Beast. “It would change the paradigm.”

Today, the research standard is to take a group of infants, give some of them a treatment and give the rest a variation or none at all, and see which do better several years down the line. By that time, research subjects have been impacted by so many things besides their time in the ICU, and it becomes difficult to know to what extent the intervention actually made difference.

“In practice, that means that people have been desperately looking for what they call proximal biomarkers—very early markers of effect of any new therapy,” Vanhatalo said. That’s what this software offers: a way to track brain development, and therefore the impacts of different interventions, in real time. If you want to know if this incubator works better than that one, or if a particular nutritional regime results in healthier babies, you can figure that out without waiting until the babies start walking and talking. “You will see the novel therapies affecting the babies within days and weeks, rather than within years.”

Today, interpreting EEGs takes time and specialized knowledge, which makes it an impractical tool in many hospitals. And even the most highly trained neurophysiologists inevitably introduce some subjectivity in their readings. “The limitation is that we don’t have really quantitative ways of assessing function; it’s all based on eyeballing it. We have criteria; we have scores; we have things that we use. But it’s still very manual and very laborious,” Cecil Hahn, a neurologist at the Hospital for Sick Children in Toronto, said. (Hahn collaborates with the BABA Center but was not involved in this specific research.)

If EEG devices came preloaded software that could automatically interpret the signal, that would make it easier for more treatment centers to use the technology—and have confidence in it. “You may not even need a neurophysiologist or EEG reader to look at the EEG, if you developed a computer-based system, and that could be deployed widely,” Hahn said.