(M-041) Integrating Machine Learning and Magnetic Resonance Imaging Radiomics for Trajectory Modeling in Duchenne Muscular Dystrophy: Explainable Artificial Intelligence to Inform Clinical Trial Design and Personalized Intervention

Hyunjae Jeong – Department of Pharmaceutics – Center for Pharmacometrics and Systems Pharmacology, Intelligent Critical Care Center, University of Florida; Rebecca Wilcocks – Department of Physical Therapy – University of Florida; Glenn Walter – Department of Physiology and Aging – University of Florida; Robert Mueller – Department of Physical Therapy – University of Florida; William Triplett – Department of Physical Therapy – University of Florida; Krista Vandenborne – Department of Physical Therapy – University of Florida; Sarah Kim – Department of Pharmaceutics – Center for Pharmacometrics and Systems Pharmacology, Intelligent Critical Care Center, University of Florida

Objectives: Duchenne muscular dystrophy (DMD) is a rare pediatric condition characterized by progressive muscle degeneration. This study aimed to inform subgroup selection for clinical trial designs and personalized interventions by quantifying the individual-level rate and severity of DMD progression trajectories using machine learning (ML) models. We leveraged previously developed disease progression models [1] and magnetic resonance imaging (MRI) derived radiomic features collected at screening visits.

Methods: This study utilized a subset of data from the ImagingNMD study (NCT01484678), which included 48 individuals with DMD assessed using Dixon MRI sequences. Radiomic features were extracted using PyRadiomics [2] from segmented soleus in 3 axial slices around the landmark slice, generating 1,130 features per individual, including shape-based, first-order, and texture-based characteristics. Baseline functional test assessments and demographics were also incorporated as inputs. Feature pre-selection was performed using Pearson correlation to remove collinear covariates, followed by the LASSO regularization to remove coefficients below 0.03. Multiple ML models were then trained and compared with 5-fold grid search cross-validation to evaluate their predictive performance for individual parameters describing the increase in fat fraction in the soleus as DMD progresses: DPT50 (age at which the measure is half of its maximum increase), DPmax (extent), γ (rate), and S0 (extrapolated measure when age is 0). Key radiomic biomarkers were identified through impurity-based feature importance and post-statistical analysis. Unsupervised clustering of key radiomic biomarkers was further applied to derive distinct subgroups.

Results: Fifty-nine baseline features (38 texture-based, 14 first-order, 4 functional assessments, BMI, age, and steroid usage) were pre-selected to train ML models. The tree-based XGBoost regression model demonstrated the best performance in predicting individualized DMD progression parameters. In addition, these evaluations fall within the typical 6-month to 1-year interval of clinical screening for individuals with DMD, with a mean absolute error of ±0.64 years for DPT50 (R²=0.79), emphasizing the model’s relevance to standard monitoring practices. Feature importance analysis revealed potential radiomic biomarkers including texture-based nonuniformity, emphasis, gray-level correlation, as well as first-order mean and interquartile range. Unsupervised clustering identified distinct subgroups with mean values of DPT50=11.66 and 15.01 years and γ=8.196 and 7.857 in fast (n=17) and slow (n=30) disease progression, respectively.

Conclusions: These findings highlight the potential of ML approaches with baseline MRI-derived radiomic features to accurately predict disease progression trajectories in individuals with DMD. The AI-assisted feature importance analysis effectively identified potential imaging biomarkers, based on their strong association with disease progression. Clustering analysis revealed clinically meaningful subgroups with interpretable decision criteria that may inform clinical trial designs and personalized intervention strategies.

Citations: [1] Kim S., Willcocks R.J., Daniels M.J., Morales J.D., Yoon D.Y., Triplett W.T., Barnard A.M., Conrado D.J., Aggarwal V., Belfiore-Oshan R., Martinez T.N., Rooney W.D., Vandenborne K. Multivariate modeling of magnetic resonance biomarkers and clinical outcome measures for Duchenne muscular dystrophy clinical trials. CPT Pharmacometrics Systems Pharmacology. 2023; 12(10): 1437-1449.

[2] van Griethuysen J. J. M., Fedorov A., Parmar C., Hosny A., Aucoin N., Narayan V., Beets-Tan R. G. H., Fillon-Robin J. C., Pieper S., Aerts H. J. W. L. Computational Radiomics System to Decode the Radiographic Phenotype. 2017, Cancer Research, 77(21), e104–e107.

Keywords: Disease Progression Modeling, Machine Learning, Radiomic Biomarkers