MR Imaging-Based Biomarkers for Strength Prediction: A Statistical Shape and Architecture Modeling of Quadriceps Muscles.

BACKGROUND

Muscle mass decline, associated with strength decline, is a hallmark of aging. Yet, strength decline greatly exceeds mass decline. This indicates that aspects of muscle quality and architecture-not reflected by mass-also influence force generating capacity. Additionally, shape modeling enables analysis of the shape variations of muscles beyond size.

PURPOSE: To predict muscle strength using muscle features beyond muscle quantity.

STUDY TYPE: Retrospective cross-sectional study.

POPULATION: Twenty-four healthy subjects normally distributed over an age range between 30 and 79 years old with a balanced sex distribution (12 female).

FIELD STRENGTH/SEQUENCE: 3 T MRI using multi-echo Dixon and Stejskal-Tanner DTI.

ASSESSMENT: Shape-only and shape + architecture models were generated using water-only and DTI images of the quadriceps. Multiple linear mixed-effects models were produced using (1) volume, (2) shape-only, and (3) shape + architecture. Volume was not added to the shape-only and shape + architecture models. Features reaching statistical significance within the models were retained for further analysis. Models' performance was evaluated using leave-one-subject-out (LOSO) cross-validation (CV).

STATISTICAL TESTS: Pairwise, subject-level bootstrapping comparison was conducted and ∆R 2 and ∆RMSE with 95% confidence interval (CI) were calculated. The improvement was considered statistically significant when both ∆R 2 and ∆RMSE are positive and the 95% CI did not contain zero. Positive ∆R 2 and ∆RMSE indicate an increase in R 2 and a decrease in RMSE values.

RESULTS

Shape-only features demonstrated an improvement in the model performance compared to muscle volume. Models were significantly improved for the vastus lateralis to predict eccentric torque-∆R 2 = 0.16 (0.01-0.29), ∆RMSE = 5.0 (0.4-9.7); and for the vastus intermedius predicting isometric torque-∆R 2 = 0.19 (0.02-0.36), ∆RMSE = 6.5 (0.7-12.0). Shape + architecture features did not significantly improve the performance (all p ≥ 0.131).

DATA

CONCLUSION

Shape-only models are promising to quantify variations of muscle shape related to force production, and have the potential to develop imaging-based biomarkers for muscle strength in diseases.

EVIDENCE LEVEL: 3.

TECHNICAL EFFICACY: Stage 2.