A statistical shape model of the human sternum including cortical bone thickness

Understanding population-level variation in bone size and shape is important for general biomechanics, clinical assessments, and for developing heterogeneous human body models used in safety restraint design. Thoracic injuries are common in severe motor vehicle crashes, and although rib fractures are frequently studied, sternal fractures also occur in these scenarios. This study aimed to investigate how demographic factors-sex, age, stature, and BMI-relate to variations in sternal size, shape, and cortical thickness, all important factors for fracture tolerance.

Computed tomography scans from 56 female and 58 male subjects were segmented using a machine-learning-based pipeline to extract sternal geometries. A template mesh was fitted to each subject using landmarking and surface-fitting techniques, and cortical bone mapping was applied to estimate cortical thickness. Principal component analysis (PCA) was then used to quantify variation in shape and thickness. Multivariate linear regression models were created for each principal component (PC) using the demographic predictors and first-order interaction terms.

The leading principal components captured variation in cortical thickness distribution, overall size, the manubriosternal joint, and the positions of the sternocostal joints. There were significant demographic trends for five PCs, which explained 39.4% of the total variance in the PCA. When using the significant regression trends to predict sternal shape and cortical thickness in the original sample, R ²-values were 0.22 for spatial coordinates and 0.07 for cortical thickness, indicating substantial unexplained individual variation. The mean sternum had an average cortical thickness of 0.92 mm.

Predictor-isolation analyses indicated that sex exerted a stronger influence on overall size than stature. Further, sternal width increased with age, and individuals with higher BMI tended to exhibit slightly greater cortical thickness. While regression models were developed to explore demographic associations, the predictive performance was limited, indicating that the model should primarily be interpreted as a descriptive representation of population-level variation rather than a predictive tool.