Magnetic resonance imaging based radiomic feature extraction of bone quality in corticosteroid treated boys with Duchenne muscular dystrophy

Topic:

Translational Research

Poster Number: 281 T

Author(s):

Remya Kunnath Ravindranunni, MPT, University of Florida, Julian McDonald, University of Pennsylvania, Jianna Kim, University of Pennsylvania, Dylan Bellissimo-Meade, University of Pennsylvania, Rishi Aravind, University of Pennsylvania, Alyssa Rosen, University of Pennsylvania, Christiana Cottrell, BA, University of Pennsylvania, Alison Barnard, DPT, PhD, University of Florida, Krista H Elvire Vandenborne, PhD, University of Florida, Glenn Walter, PhD, University of Florida, Chamith Rajapakse, PhD, University of Pennsylvania, Rebecca Wilcocks, PhD, University of Florida

Background: In Duchenne muscular dystrophy (DMD), osteoporosis increases morbidity and mortality. Improved biomarkers of bone quality are needed for therapeutic development and clinical care. High resolution magnetic resonance imaging (MRI) is an effective non-invasive tool to assess bone microstructure which requires specialized acquisition and processing tools. Radiomic biomarkers can quantify systemic bone loss in other osteoporotic populations. This study investigates whether radiomic features extracted from standard research magnetic resonance imaging (MRI) scans can discriminate bone quality between corticosteroid-treated boys with DMD and age-matched healthy controls. Methods: In this cross-sectional study, boys with DMD (n = 21, age: 13.2 ± 3.2 years, corticosteroid use: 7.8 ± 3.4 years) and healthy controls (n = 14, age: 13.6 ± 3.9 years) underwent 3T MRI (Siemens Prisma) imaging of the distal femur using 12-echo gradient echo sequence (TR = 7.1 ms, TE1 = 1.8 ms, echo spacing = 0.7 ms). Trabecular ROIs, scaled to femur length, were analyzed using 3D Slicer [1] with the radiomics extension [2]. Sixty radiomic features were extracted. Cohen’s D and unpaired t-tests identified features that significantly differentiated bone quality between groups. Results: Of the 60 features analyzed, 33 significantly differentiated bone quality between groups (Cohen’s D ≥ 0.8, P ≤ 0.05). The most differing clinically relevant features include first-order features (10th percentile, mean absolute deviation), grey-level run length matrix features (run entropy and run length non-uniformity) and grey level dependence matrix features (dependence non-uniformity). Collectively, these features capture loss of low-intensity pixels that are comprised primarily of trabecular bone in DMD, as well as changes in the trabecular bone structure. Conclusion: Radiomic features from standard MRI can effectively differentiate bone quality in DMD Vs controls. These findings support the potential of MR-based radiomics in identifying bone biomarkers in DMD. Funding Sources: US Department of Defense (W81XWH-21-1-0566, PI: Willcocks/Rajapakse).