A machine-learning random survival forest model integrating quantitative MRI fat fraction metrics and clinical variables showed favourable predictive performance for time to loss of ambulation in boys with DMD across short- and long-term horizons, and was deployed as an online tool to support clinical decision-making and trial enrichment.
In a prospective cohort of 188 boys with DMD, 35.1% lost ambulation over a median 43.5 months of follow-up.
Feature selection retained six predictors of time to LOA: two clinical function measures, age, and three muscle fat fraction parameters.
The RSF model had the highest discrimination in the test set, outperforming the clinical-only Cox model and the ridge Cox model.
The RSF model's time-dependent AUC was high at all horizons but declined with longer prediction time.
RSF showed the best calibration, with the lowest integrated Brier score.
The RSF model stratified patients into low-, intermediate- and high-risk groups with markedly different LOA event rates.
SHAP analysis identified exploratory thresholds for key predictors associated with higher LOA risk.
The model was deployed as an interactive web application for individualized LOA risk prediction.
Duchenne muscular dystrophy (DMD) is a muscle-wasting disease in which boys eventually lose the ability to walk, but the timing varies a lot from child to child. This makes it hard for families and doctors to plan, and hard to design clinical trials. This study followed 188 boys with DMD for a median of about 3.6 years, using MRI scans that measure how much fat has replaced muscle in hip and thigh muscles, together with standard clinical measures such as walking-function scores, strength and age. The researchers built a machine-learning model (a random survival forest) that used six factors: a walking-function score (NSAA), a muscle strength measure (MRC%), age, and fat content in three muscles. In a held-out test group of 40 boys, it predicted who would lose the ability to walk better than a model using clinical information alone, and it was especially accurate for predictions within 1 to 2 years. It also sorted boys into low-, intermediate- and high-risk groups; none of the low-risk group lost ambulation during follow-up, while about 35% per year of the high-risk group did. An online tool was created to generate individual risk estimates. This research suggests that adding MRI measures of muscle fat to clinical information could help forecast disease course and possibly select patients for trials. Caveats are that the test set was small (40 boys, 15 events), the threshold values identified are described as exploratory, and the model would need confirmation in other groups of patients before wider use.
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Upload Your LabsLiu C, Song Y, Wan J, Xu T, Zhang X, Yan Y, et al.. (2026). Machine Learning Integrating Quantitative MRI and Clinical Features Predicts Loss of Ambulation in Duchenne Muscular Dystrophy.. Journal of cachexia, sarcopenia and muscle. https://doi.org/10.1002/jcsm.70413