Eye Scans May Help to Diagnose MS in Children

Researchers used machine learning from optical coherence tomography (OCT) scans

Margarida Maia, PhD avatar

by Margarida Maia, PhD |

Share this article:

Share article via email
An enlarged image of the human eye is captured in the lens of a giant telescope as a person looks at the stars.

A machine learning approach based on eye scans was employed by researchers to diagnose multiple sclerosis (MS) in children with up to 80% accuracy.

Optical coherence tomography (OCT) scans also provided enough data to diagnose other demyelinating diseases with 75% accuracy. OCT is an imaging tool that uses light waves to take pictures of the retina — a layer of light-sensitive nerve cells found at the back of the eye that plays a central role in vision — with the help of a computer.

Moreover, researchers found that some imaging features may help distinguish MS from other demyelinating diseases. These “specific anatomic areas that may be of high utility in differentiating MS from other disorders,” they wrote.

The study, “Machine learning classification of multiple sclerosis in children using optical coherence tomography,” was published in the Multiple Sclerosis Journal.

MS is a demyelinating disease that occurs when the immune system attacks myelin — a fatty substance that sheathes nerve cells — and ultimately damages nerve cells themselves. This can bring about a wide range of symptoms that sometimes overlap those of other demyelinating diseases.

Recommended Reading
pediatric MS | Multiple Sclerosis News Today | Aubagio as pediatric therapy | illustration of adult hugging child

Gilenya, Tysabri Show Superiority at Preventing Relapses in Children

While most patients experience the first symptoms in their 20s and up to their 50s, some may develop the disease earlier during childhood or adolescence. But a timely diagnosis isn’t always easy.

Machine learning is a form of artificial intelligence that allows a computer to learn by scanning through large collections of data. An algorithm, which is a set of step-by-step instructions for accomplishing a specific task or solving a particular problem, is then ran to help the computer find patterns in the data.

In the clinic, this could be used to examine medical images and aid in the diagnosis of certain diseases. For this to happen, however, scientists first would have to determine which data features are best for the computer to learn. What’s more, there are many algorithms, and choosing the correct one is key to obtain reliable results.

Here, a team of researchers in Canada used machine learning to diagnose MS in children. Researchers already knew that children with MS most often have vision problems, so they drew on data from OCT scans.

The study included 187 children with demyelinating diseases; 92 had monophasic acquired demyelinating syndrome, 57 had MS, 27 had myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD), and 11 had neuromyelitis optica spectrum disorder (NMOSD). It also included 69 healthy children who served as controls.

All children had both eyes scanned by OCT. Scans were taken in a dark room without eye tracking, meaning the scanner did not track the naturally moving eye.

Data from a total of 512 eyes (374 from patients and 138 from controls) were included in the analyses. Patients were a mean 1.9 years younger than controls when they had the scan (12.8 vs. 14.7 years). Those with MS had the disease for a mean of 7.2 months. Other demyelinating diseases lasted for longer (up to a mean of two years).

Looking for patterns

To look for patterns in the data obtained from the scans, researchers put machine learning into action. They ran and compared 10 different algorithms for their ability to accurately distinguish MS patients from those with other demyelinating diseases or healthy children. An algorithm called “random forest” worked best and was used in all later analyses.

As it read through data from 24 different OCT features, the random forest algorithm was able to tell patients with demyelinating diseases from controls with 72% accuracy. The accuracy was best when only 15 of the 24 features were taken into account.

When researchers focused on MS, they found that eight features were enough to achieve an accuracy of 80% in telling patients from controls. Moreover, it was possible to tell MS from other demyelinating diseases with 68%  accuracy.

MRI is the gold standard imaging tool in the diagnosis of MS. But these findings suggest that “OCT features may be as useful and readily clinically applicable in diagnosing MS in children without any a priori criteria,” the researchers wrote, adding that more research is needed.