Abstract
Introduction & Purpose
Femoral neck–shaft angle (NSA) and femoral anteversion angle (AVA) influence lower-limb
alignment, joint loading, and movement biomechanics during growth (Figure 1a). Altered
femoral geometry has been associated with changes in gait mechanics and musculoskeletal
overload, highlighting its relevance for clinical biomechanics, sports medicine, and paediatric
movement research (Kainz et al., 2023). Although femoral morphology has been widely
investigated (Szuper et al., 2015; Staheli et al., 1985; Bobroff et al., 1999), large-scale
assessment remains challenging due to time-consuming manual measurements, operator
dependency, and methodological variability (Boese et al., 2016; Kaiser et al., 2016). This study
aimed to develop an automated MRI-based workflow combining deep learning femur
segmentation with deterministic geometric analysis to quantify NSA and AVA and generate
reference distributions in typically developing children.
Methods
Whole-limb MRI scans of 52 typically developing children, 24 female and 28 male, without
known lower-limb pathologies or pain, aged 5–16 years, were analysed. In total, 104 femora
were available. Femora for which automated geometric fitting did not yield plausible angle
estimates were excluded. Femoral segmentation was performed using a self-trained 3D nnU-
Net model (Isensee et al., 2021) trained on 37 manually annotated femur segmentations
(Figure 1b). The resulting masks were processed using the deterministic MSKsInertiaGeom
pipeline, combining STAPLE-based processing of segmented bone geometries, following
Modenese and Renault (2021), with inertia-axis calculations and refined geometric fitting to
derive NSA and AVA without manual landmark placement. Linear mixed-effects models
accounted for bilateral femoral measurements, with participant identity as a random effect and
age, sex, and age-by-sex interaction as fixed effects. Significance was defined as p < 0.05.
Results
Automated geometric fitting succeeded in 102 of 104 femora. Mean NSA was 132.9 ± 5.8° with
a range of 120° to 145°, while mean AVA was 22.4 ± 10.0° with a range of 0° to 45°. Linear
mixed-effects modelling demonstrated a significant negative association between age and
NSA, with β = −0.94°/year and p < 0.001. AVA showed no statistically significant association
with age, with β = −0.70°/year and p = 0.109. No statistically significant sex-related differences
were observed for NSA or AVA, and no significant age-by-sex interactions were detected. The
resulting distributions provide reference values across the investigated age range (Figure 1c).
Discussion
This automated MRI-based framework enables scalable, operator-independent quantification
of paediatric femoral morphology. The observed age-related decrease in NSA is consistent
with reported developmental patterns of proximal femoral morphology (Hefti, 2007), whereas
AVA demonstrated substantial inter-individual variability and no significant age dependency.
Limitations include the single-centre and cross-sectional design, lack of formal validation
against independent manual segmentation and measurement protocols and failed geometric
fitting in two femora.
Conclusion
An automated MRI-based pipeline integrating deep learning segmentation and deterministic
geometric analysis enables objective quantification of paediatric femoral NSA and AVA. These
reference values may support future biomechanical and sports science research investigating
femoral morphology in movement development, physical activity, sports participation, and
musculoskeletal overload during growth.
References
1. Kainz, H., Mindler, G. T., & Kranzl, A. (2023). Influence of femoral anteversion angle and
neck-shaft angle on muscle forces and joint loading during walking. PLOS ONE, 18(10),
e0291458. https://doi.org/10.1371/journal.pone.0291458
2. Szuper, K., Schlégl, Á. T., Leidecker, E., Vermes, C., Sándor, B., & Than, P. (2015). Three-
dimensional quantitative analysis of the proximal femur and the pelvis in children and
adolescents using an upright biplanar slot-scanning X-ray system. Pediatric Radiology, 45,
411–421. https://doi.org/10.1007/s00247-014-3146-2
3. Staheli, L. T., Corbett, M., Wyss, C., & King, H. (1985). Lower-extremity rotational problems
in children. The Journal of Bone and Joint Surgery. American Volume, 67(1), 39–47.
https://doi.org/10.2106/00004623-198567010-00006
4. Bobroff, E. D., Chambers, H. G., Sartoris, D. J., Wyatt, M. P., & Sutherland, D. H. (1999).
Femoral anteversion and neck-shaft angle in children with cerebral palsy. Clinical
Orthopaedics and Related Research, 364, 194–204. https://doi.org/10.1097/00003086-
199907000-00025
5. Boese, C. K., Dargel, J., Oppermann, J., Eysel, P., Scheyerer, M. J., Bredow, J., & Lechler,
P. (2016). The femoral neck-shaft angle on plain radiographs: A systematic review.
Skeletal Radiology, 45, 19–28. https://doi.org/10.1007/s00256-015-2236-z
6. Kaiser, P., Attal, R., Kammerer, M., et al. (2016). Significant differences in femoral torsion
values depending on the CT measurement technique. Archives of Orthopaedic and
Trauma Surgery, 136, 1259–1264. https://doi.org/10.1007/s00402-016-2536-3
7. Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-
Net: A self-configuring method for deep learning-based biomedical image segmentation.
Nature Methods, 18, 203–211. https://doi.org/10.1038/s41592-020-01008-z
8. Modenese, L., & Renault, J.-B. (2021). Automatic generation of personalised skeletal
models of the lower limb from three-dimensional bone geometries. Journal of
Biomechanics, 116, 110186. https://doi.org/10.1016/j.jbiomech.2020.110186
9. Hefti, F. (2007). Pediatric Orthopedics in Practice. Springer. https://doi.org/10.1007/978-3-
540-69964-4