Using AI to detect cerebral palsy in babies

Lars Adde, a specialist in pediatric physiotherapy and researcher at St. Olavs hospital in Trondheim, is developing an AI tool to detect cerebral palsy (CP) in babies. He is featured in the Teknisk Ukeblad technology podcast "Teknisk sett." He hopes hospitals can start using the invention in three to four years.

The idea rests on the fact that small babies wave their arms and legs in all directions, and that after a few months the movements become more deliberate. A trained eye can spot deviations in these movements, which can signal conditions such as CP. Adde was trained to see these patterns more than 20 years ago, but he is one of few clinicians with that skill, and many hospitals lack such specialists. He therefore wanted to measure the movement patterns objectively. CP is brain damage that impairs motor function and the ability to control and coordinate movement. About two in 1,000 newborns are diagnosed with it, and for babies in neonatal intensive care because of extremely premature birth or other medical risk factors, the figure is up to 8-10 percent.

Working with NTNU's department of engineering cybernetics, the team first placed babies on a mattress with sensors on their arms and legs and tracked positions in a magnetic field. This used traditional statistics and machine learning, but it required lab equipment, and attaching sensors and wires to infants made good measurements difficult.

Digital video solved this. Babies can be filmed lying on their backs and kicking, with no sensors or discomfort. The researchers collected 20,000 images of infants and marked wrists, knees, hips and noses, then trained a model to find these points in new videos. The model now tracks 19 points on a baby and recognizes the movement patterns of children with CP about as accurately as Adde himself.

Getting to clinical use will take time and funding. Adde has become an entrepreneur and has secured financing for the next two to three years. He hopes the model will free specialists' time for children who are sick, arguing that healthcare must use digital tools more and become more efficient.