A study in Clinical Pharmacology & Therapeutics reports that machine learning models fed with baseline MRI radiomics can estimate how fast individual patients with Duchenne muscular dystrophy (DMD) will progress, and can sort them into slow and fast subgroups for trial design1.
What the study found
The researchers extracted radiomic features from fat-fraction maps of two leg muscles, the soleus and the vastus lateralis, producing 810 features for each patient. After feature selection they added functional and demographic measures as covariates1.
Several machine learning regressors were then trained and validated to estimate the parameters that describe each patient's MR-spectroscopy fat-fraction trajectory over time. These were DPT 50 (the age at half of the maximal increase), DP max (the extent of change), γ (the rate) and S 0 (the value extrapolated to age 0)1.
A tree-based XGBoost model performed best at forecasting future longitudinal fat-fraction trajectories. In the soleus it reached a mean absolute error of 0.04 and an R2 of 0.94, and in the vastus lateralis a mean absolute error of 0.06 and an R2 of 0.951.
Unsupervised clustering then split patients into slow-progression and fast-progression groups with interpretable cut-offs. In the soleus, DPT 50 was 12.21 in one group against 14.83 in the other, with γ of 8.07 against 7.54. In the vastus lateralis the DPT 50 values were 10.37 and 14.21, with γ of 7.72 and 7.091. Feature importance analyses also pointed to candidate biomarkers that the authors link clinically to disease progression, and a patient-level reading of those features suggested scope for tailoring interventions to the individual1.
What it means for the market
The commercial relevance sits with sponsors running DMD programmes. Variable progression between patients makes treatment effects harder to detect, and the authors argue that grouping patients by predicted trajectory at baseline can inform trial designs and allow therapeutic effects to be assessed within more homogeneous groups1. If the approach holds up, a baseline MRI scan could support enrichment or stratification decisions before dosing, rather than relying on observed decline during the study.
The paper frames the work as a step towards AI-assisted clinical trial development and MRI-guided disease modelling in neuromuscular disorders1. For companies building imaging biomarkers or trial-optimisation software, it adds published evidence that radiomic inputs carry predictive signal in a rare paediatric disease, an area where small patient numbers make every enrolled participant expensive.
What the published abstract does not settle is how far the models generalise. It gives no figure for the number of patients or sites, does not describe external validation in an independent cohort, and does not report use in a live trial1. Regulators would also need to accept MRI-derived subgroups before sponsors could rely on them for registrational studies, and the source does not address that question.
The authors present the patient-level interpretation as showing only the possibility of personalised intervention strategies, so the clinical benefit of acting on these predictions remains untested1.
Sources
- Advancing Clinical Trials and Precision Medicine in Duchenne Muscular Dystrophy Using Machine Learning-Based Trajectory Modeling and MRI Radiomics. (opens in a new tab)Clinical pharmacology and therapeutics · doi.org
This briefing summarises publicly available research and reporting for information only. It is not medical, investment or legal advice. Follow the sources listed for the primary research and reporting.