Abstract
Proximal femoral medullary canal morphology is a key determinant of cementless stem fit, primary stability, and load transfer in total hip arthroplasty (THA), yet population-level three-dimensional characterization remains limited. This study was designed to quantify variability in canal geometry and to capture dominant anatomical modes of variation with statistical shape modeling (SSM). Computed tomography data from 763 candidates for primary THA (389 female, 374 male; 20-92 years) were analyzed. Endosteal contours of the proximal canal were processed to build a point-correspondent SSM by principal component analysis (PCA). Five geometric features were evaluated per specimen: equivalent radius (normalized), roundness, major-axis angle (torsion), flare index, and curvature. Substantial inter-individual variability was observed across all features, with differences by sex and age. The first three principal components accounted for 68.4% of total variance, and each showed interpretable associations with at least one geometric feature. Model behavior was examined by synthetic sampling within ±2 SD (specificity) and by 10-fold cross-validation (generalization), indicating faithful reconstruction of real shapes and stable performance on held-out data. These findings provide a compact description of proximal canal shape variation and its key geometric drivers. The resulting population map is expected to support implant selection and sizing in preoperative planning, inform shape-based classification, and guide design envelopes for standard and personalized stems, with potential efficiencies in manufacturing and material use.