New AI-based tool may predict when Duchenne kids will lose ability to walk
DMD-TIMER aims to personalize treatment decision-making for children
Written by |
Scientists have developed a machine-learning tool to predict the loss of walking ability in children with Duchenne MD. (Image from iStock)
Using a computer-based approach, researchers in China have created an online tool, dubbed DMD-TIMER, to predict when children with Duchenne muscular dystrophy (DMD) are likely to lose the ability to walk.
The new tool, which uses clinical and MRI assessments, was developed using machine learning, a type of artificial intelligence (AI) that employs algorithms to learn from data and make predictions. The goal, according to the team, is to better predict when youngsters with the genetic disease, which causes progressive muscle weakness and wasting, will face “loss of ambulation” — a “key milestone” in DMD care and treatment that nonetheless has “highly variable timing.”
The results of a new study testing the tool “underscore the potential of DMD-TIMER for personalized clinical decision-making across short- and long-term prediction horizons,” the researchers wrote.
Specifically, it will allow clinicians to assess a patient’s risk of losing walking ability over the following four years. It works by entering the outcomes of six key clinical and MRI tests, the team noted.
“Early [loss of ambulation] prediction could inform timely clinical decisions, … potentially prolonging ambulation and improving quality of life,” the researchers wrote.
Details of the tool’s development were described in “Machine Learning Integrating Quantitative MRI and Clinical Features Predicts Loss of Ambulation in Duchenne Muscular Dystrophy,” a study published in the Journal of Cachexia, Sarcopenia and Muscle.
In DMD, a deficiency in dystrophin — a muscle protein that acts like a shock absorber during body movement — causes muscle cells to break down and be gradually replaced by fat and scar tissue. This leads to muscle weakness and, eventually, loss of ambulation, or the inability to walk independently.
To measure walking ability, clinicians often use tools such as the North Star Ambulatory Assessment (NSAA), a motor function rating scale that depends on effort and cooperation. Such tests can be difficult, especially for some children. MRI scans can also help, as they measure fat infiltration in muscle without surgery or needles.
Still, few studies have combined MRI scans with clinical assessments to predict when DMD patients will lose ambulation.
Tool to predict walking ability nearly perfect at 1 year
To improve such predictions, researchers from Sichuan University in Chengdu turned to AI and a computer-based approach, developing a machine-learning tool that could learn from MRI and clinical data and better gauge the expected time to loss of ambulation for children with DMD.
Their study enrolled 188 boys with Duchenne, who had a median age of 9.1 years. Over a median follow-up of 43.5 months, or nearly four years, slightly more than one-third lost ambulation.
From patient data, the team identified six predictors significantly associated with the loss of walking ability: age, NSAA score, muscle strength assessments, and fat fraction of three leg and hip muscles. Fat fraction is the proportion of the MRI signal that comes from fat. Muscle strength was measured by the Medical Research Council scale (MRC%).
To assess how well these combined data predicted loss of ambulation over time, the researchers used a tool called area under the curve (AUC), with values closer to 1 indicating better prediction. The model’s predictive value was nearly perfect at one year, with an AUC of 0.984. The values then gradually declined, with AUCs of 0.971 at two years, 0.913 at three years, and 0.850 at four years.
Based on loss of ambulation, the team then divided patients into low-, intermediate-, and high-risk groups. Annual loss-of-ambulation rates were 7.2% in the intermediate-risk group and 34.6% in the high-risk group, with no losses in the low-risk group.
[The new model] exhibited favourable predictive performance for time to [loss of ambulation] in DMD across short- and long-term horizons.
When the researchers estimated how much each factor contributed to an individual prediction, the NSAA score was the most influential. This was followed, in order of importance, by the three fat fraction measures, MRC%, and then age. Over time, the fat fractions became more important, while the NSAA score became less influential, the researchers noted
The analysis also identified test results associated with a higher risk of ambulatory loss. These included an NSAA score lower than 14.55 and an MRC% lower than 68.36%. Fat fractions higher than 32.3% for the rectus femoris and 58% for vastus lateralis, both muscles in the thigh, were also linked to greater risk, as were results higher than 7.43% for the obturator internus, a hip muscle.
According to the team, their model “exhibited favourable predictive performance for time to [loss of ambulation] in DMD across short- and long-term horizons,” and showed potential for clinical use.
“We successfully developed and validated a machine learning-based risk stratification tool that integrates clinical variables and quantitative MRI fat fraction parameters to predict time to [loss of ambulation] in DMD,” the researchers concluded.
Leave a comment
Fill in the required fields to post. Your email address will not be published.