Predicting the Biomechanical Strength of Proximal Femur Specimens through High Dimensional Geometric Features and Support Vector Regression.

Abstract

Estimating local trabecular bone quality for purposes of femoral bone strength prediction is important for improving the clinical assessment of osteoporotic hip fracture risk. In this study, we explore the ability of geometric features derived from the Scaling Index Method (SIM) in predicting the biomechanical strength of proximal femur specimens as visualized on multi-detector computed tomography (MDCT) images. MDCT scans were acquired for 50 proximal femur specimens harvested from human cadavers. An automated volume of interest (VOI)-fitting algorithm was used to define a consistent volume in the femoral head of each specimen. In these VOIs, the non-linear micro-structure of the trabecular bone was characterized by statistical moments of its BMD distribution and by local scaling properties derived from SIM. Linear multi-regression analysis and support vector regression with a linear kernel (SVRlin) were used to predict the failure load (FL) from the feature sets; the predicted FL was compared to the FL values determined through biomechanical testing. The prediction performance was measured by the root mean square error (RMSE) for each image feature on independent test set. The best prediction result was obtained from the SIM feature set with SVRlin, which had the lowest prediction error (RMSE = 0.842 ± 0.209) and which was significantly lower than the conventionally used mean BMD (RMSE = 1.103 ± 0.262,, p<0.005). Our results indicate that the biomechanical strength prediction can be significantly improved in proximal femur specimens on MDCT images by using high-dimensional geometric features derived from SIM with support vector regression.