Development of an MRI-based Automatic Segmentation Model to analyze Sex-Related Differences in the Lower-extremity Musculoskeletal System

Abstract

Despite the recognized impact of sex on biomechanics, research remains biased toward male anatomy, raising concerns about the validity of musculoskeletal (MSK) model predictions for females. This study investigated whether sex-specific bone geometry variations predict differences in the proportional volumes of the Gluteus Maximus (GMAX) and Rectus Femoris (RFEM). Using an MRI-based nnU-Net segmentation model trained in this thesis, bone metrics were extracted, and muscle volumes were normalized to derive proportional volumes for 16 young adults (9F/7M). The segmentation model demonstrated high accuracy (DSC: 0.926 for bones, 0.954 for muscles), revealing significant sexual dimorphism in bone geometry. Males exhibited greater femoral offsets and knee widths, while females had larger posterior pelvic widths and depths. %RFEM was significantly higher in males (p = 0.01), but %GMAX showed no sex-related differences. Regression analysis identified femoral offset and femur length as partial predictors of %RFEM (R^2 = 0.478), with pelvis-femur length weakly predicting %GMAX (R^2 = 0.151). However, the low predictive power suggests limitations in using bone metrics to estimate muscle volume proportions. These findings indicate that femoral dimorphism may partially explain sex-related %RFEM differences, but its role in %GMAX remains unclear. Future research integrating additional biomechanical factors could enhance sex-specific MSK modeling accuracy.