New MRI approach using AI may offer better picture of FA changes
Study shows automated MRI analysis can track progression over four years
MRI scans analyzed with artificial intelligence (AI) to measure the size of the dentate nucleus — a deep brain region that tends to shrink in people with Friedreich’s ataxia — may help track how the disease progresses over time, according to a new study.
“These volumetric changes were associated with longer disease duration and worsening clinical scores … highlighting the potential utility of [dentate nucleus] volume as a neuroimaging marker for disease progression,” the researchers wrote.
The study, “Deep learning-based 3D reconstruction of dentate nuclei in Friedreich’s ataxia from T2*weighted MR images,” was published in Machine Learning with Applications.
How FA affects the brain and why the dentate nucleus matters
Muscle weakness in FA often results from damage to the spinal cord and the cerebellum, a coordination center at the back of the brain. One of the most affected areas is the dentate nucleus (DN), an iron-rich structure deep within the cerebellum. Because studies of brain tissue have shown major loss of DN volume in people with FA, measuring this structure over time may help track disease progression.
However, standard MRI scans don’t clearly show the DN because its high iron content interferes with the image. A different MRI measure, called the T2* relaxation rate, is more useful since it is strongly affected by iron and helps bring this structure into focus.
Currently, the gold standard for examining brain structures relies on manual segmentation, where a trained expert traces the outline of the structure by hand to calculate its size. This method is extremely time-consuming and can produce different results depending on who performs it.
In recent years, machine learning — a form of AI that learns patterns from imaging data — has been used to automate this segmentation process. These tools can offer faster and more consistent alternatives to the labor-intensive manual approach.
AI tools offer a new way to measure dentate nucleus volume
A research team in Germany has developed a machine learning method that can measure DN volumes using just a few T2*-weighted MRI scans. The team then applied this tool to people with FA and healthy volunteers to track DN volume over as long as four years.
In total, the study included 33 people with FA and 33 healthy, age- and sex-matched controls. The FA group was enrolled as part of the European Friedreich’s Ataxia Consortium for Translational Studies (EFACTS).
To check how accurate the approach was, the researchers used the Dice score, a measure of how closely two sets of results match. In this study, it compared DN volume measured by the AI tool with measurements done by hand. Dice scores range from 0 to 1, with 1 meaning the two measurements match perfectly.
The initial analysis showed average Dice scores of 0.76 for people with FA and 0.81 for healthy controls using the machine learning approach.
The researchers then pretrained the machine learning model using Fastsurfer, a tool that automates the processing of brain MRI scans. After this step, Dice scores rose to 0.83 in people with FA (about a 10% increase) and 0.85 in controls (about a 5% increase).
What MRI volume changes showed about FA progression
The team then compared DN volume changes over time in people with FA and in healthy controls.
At the first MRI (baseline), DN volume was already significantly smaller in people with FA than in controls — 1,062 vs. 1,261 cubic millimeters, mm3 — even after adjusting for head size. These differences stayed consistent across follow-up visits.
Despite these differences, starting DN volume did not correlate with disease severity, as measured by the Scale for the Assessment and Rating of Ataxia (SARA).
Still, patients who started with smaller DN volumes showed larger increases in their ADL scores over the next four years — indicating greater functional decline.
Over the study period, right DN volume shrank more in people who had lived with FA longer — about 1.5 mm3 loss per year. Worsening SARA and ADL scores were also associated with additional right DN volume loss of about 2-3 mm3.
“This study demonstrates the feasibility and clinical relevance of [a machine learning]-based approach for segmenting the DN using commonly acquired T2*-weighted MRI, even with limited data availability,” the researchers wrote. “Volumetric analysis of DN revealed significantly reduced DN volumes in [FA] patients compared to controls, with reductions correlating with disease duration and progression of clinical severity over time.”
Further validation of this study’s results in a larger data set is needed, the team wrote.
About the Author
