Machine learning used to estimate when brain shrinkage begins

On average, that happens more than 5 years before symptoms appear, study finds

Marisa Wexler, MS avatar

by Marisa Wexler, MS |

Share this article:

Share article via email
An illustration of a person's brain, shown in profile.

Brain atrophy (shrinkage) in people with multiple sclerosis (MS) begins on average more than five years before disease symptoms appear, according to a new study based on machine learning models.

“Although the onset of progressive brain tissue loss measured by MRI is not synonymous with the true biological disease onset, our results suggest a major improvement in estimating MS disease duration compared to the standard practice of defining the disease onset as the time of first clinical symptom,” researchers wrote.

These results “may have significant implications for MS clinicians, researchers, and patients, and could lead to a fundamental shift in our disease understanding and, one day, determining its cause,” they added.

Recommended Reading
A person lays on a gurney while being given an IV infusion.

MS symptoms can fluctuate with Ocrevus, but ‘wearing off’ not likely

The study, “Toward precision medicine using a ‘digital twin’ approach: modeling the onset of disease-specific brain atrophy in individuals with multiple sclerosis,” was published in Scientific Reports.

Brain atrophy occurs as a normal part of aging in all adults, but in people with MS, the brain shrinks much faster due to inflammatory damage caused by the disease. A particular part of the brain called the thalamus, which normally helps to relay signals between different areas of the brain, is especially susceptible to MS-related atrophy.

Previous studies have suggested that increased brain atrophy, and specifically thalamic atrophy, is evident in people with MS years before symptoms appear, which implies that disease-related biological processes start several years before clinical onset.

The exact timing of when MS-related brain atrophy diverges from normal age-related brain atrophy has been hard to pin down, however. In this study, a team of scientists in the U.S. set out to estimate this timing using a computational strategy called the “digital twin” approach.

How the digital twin approach works

In simplest terms, the digital twin approach involves using detailed statistical models paired with advanced machine learning in order to create a computer-based simulation of the rate of brain atrophy for a given patient. The researchers made two models of brain atrophy over time: one in the context of MS, the other in the context of healthy aging.

Then they used these models to compare when thalamic atrophy began for people with MS, compared to what might have been expected if those same individuals had experienced only age-related atrophy.

“In this study, we apply the health digital twin conceptual framework to build an [artificial intelligence] algorithm in addressing a fundamental clinical problem in multiple sclerosis, which is to identify the disease-related onset of brain atrophy,” the researchers wrote.

Average times until atrophy onset

Results showed that the average time at which MS-related atrophy diverged from normal age-related atrophy was 5.1 years prior to the onset of clinical symptoms. Looking at all the times for each individual patient in the analysis, the median time was six years before clinical onset.

The researchers stressed that these numbers are only estimates based on computational models, noting that the models themselves are limited by the data that’s available to build the models. Still, they said these estimations could help to better understand the actual biological origins of MS.

“While there is no ground truth to validate our findings, this is consistent with clinical observations that white matter lesions [and] thalamic atrophy are already present prior to first clinical symptoms in MS,” they wrote.

The team also noted that a similar digital twin approach could be used to estimate changes in brain atrophy in other neurological disorders, where biological processes also likely precede clinical symptoms by several years.