Development and Validation of Statistical Models of Femur Geometry for Use with Parametric Finite Element Models.

Abstract

Statistical models were developed that predict male and female femur geometry as functions of age, body mass index (BMI), and femur length as part of an effort to develop lower-extremity finite element models with geometries that are parametric with subject characteristics. The process for developing these models involved extracting femur geometry from clinical CT scans of 62 men and 36 women, fitting a template finite element femur mesh to the surface geometry of each patient, and then programmatically determining thickness at each nodal location. Principal component analysis was then performed on the thickness and geometry nodal coordinates, and linear regression models were developed to predict principal component scores as functions of age, BMI, and femur length. The average absolute errors in male and female external surface geometry model predictions were 4.57 and 4.23mm, and the average absolute errors in male and female thickness model predictions were 1.67 and 1.74mm. The average error in midshaft cortical bone areas between the predicted geometries and the patient geometries was 4.4%. The average error in cortical bone area between the predicted geometries and a validation set of cadaver femur geometries across 5 shaft locations was 2.9%.