Predictive statistical models of baseline variations in 3-D femoral cortex morphology

Quantifying human femoral cortex morphology is important for forensic science, surgical planning, prosthesis design and musculoskeletal modeling. Previous studies have been restricted by traditional zero or one dimensional morphometric measurements at discrete locations. We have used automatic image segmentation and statistical shape modeling methods to create predictive models of baseline 3-D femoral cortex morphology on a statistically significant population. A total of 204 femurs were automatically segmented and measured to obtain 3-D shape, whole-surface cortical thickness, and morphometric measurements. Principal components of shape and cortical thickness were correlated to anthropological data (age, sex, height and body mass) to produce predictive statistical models. We show that predictions of an individual's age, height, and sex can be improved by using 3-D shape and cortical thickness when compared with traditional morphometric measurements. We also show that femoral cortex geometry can be predicted from anthropological data combined with femoral measurements with less than 2.3 mm root mean square error, and cortical thickness with less than 0.5 mm root mean square error. The predictive models presented offer new ways to infer subject-specific 3-D femur morphology from sparse subject data for biomechanical simulations, and inversely infer subject data from femur morphology for anthropological and forensic studies.