Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2014 Feb 4;23(1):013013.
doi: 10.1117/1.JEI.23.1.013013. Epub 2014 Jan 9.

Improving Bone Strength Prediction in Human Proximal Femur Specimens Through Geometrical Characterization of Trabecular Bone Microarchitecture and Support Vector Regression

Affiliations
Free PMC article

Improving Bone Strength Prediction in Human Proximal Femur Specimens Through Geometrical Characterization of Trabecular Bone Microarchitecture and Support Vector Regression

Chien-Chun Yang et al. J Electron Imaging. .
Free PMC article

Abstract

We investigate the use of different trabecular bone descriptors and advanced machine learning tech niques to complement standard bone mineral density (BMD) measures derived from dual-energy x-ray absorptiometry (DXA) for improving clinical assessment of osteoporotic fracture risk. For this purpose, volumes of interest were extracted from the head, neck, and trochanter of 146 ex vivo proximal femur specimens on multidetector computer tomography. The trabecular bone captured was characterized with (1) statistical moments of the BMD distribution, (2) geometrical features derived from the scaling index method (SIM), and (3) morphometric parameters, such as bone fraction, trabecular thickness, etc. Feature sets comprising DXA BMD and such supplemental features were used to predict the failure load (FL) of the specimens, previously determined through biomechanical testing, with multiregression and support vector regression. Prediction performance was measured by the root mean square error (RMSE); correlation with measured FL was evaluated using the coefficient of determination R2. The best prediction performance was achieved by a combination of DXA BMD and SIM-derived geometric features derived from the femoral head (RMSE: 0.869 ± 0.121, R2: 0.68 ± 0.079), which was significantly better than DXA BMD alone (RMSE: 0.948 ± 0.119, R2: 0.61 ± 0.101) (p < 10-4). For multivariate feature sets, SVR outperformed multiregression (p < 0.05). These results suggest that supplementing standard DXA BMD measurements with sophisticated femoral trabecular bone characterization and supervised learning techniques can significantly improve biomechanical strength prediction in proximal femur specimens.

Keywords: bone mineral density; dual x-ray absorptiometry; osteoporosis; quantitative computer tomography; scaling index method; support vector regression; trabecular bone.

Figures

Fig. 1
Fig. 1
Multidetector computer tomography (MDCT) images of selected femur specimens. From left to right, the specimens are categorized as high, medium, and low, based on failure load. The osteo phantom used for each specimen is also shown at the bottom.
Fig. 2
Fig. 2
Cross-sectional regions of interest (ROIs) of the volumes of interest (VOIs) defined in the femoral head (a), neck (b), and trochanter (c). The top row shows the ROIs overlaid on MDCT images of these regions. The bottom row shows these ROIs where pixel intensities are indicative of bone mineral density (BMD, mg/cm3). Note that the neck and trochanter images are of different scales and, thus, zoomed in for purposes of presentation only.
Fig. 3
Fig. 3
ROIs with BMD values extracted from the head VOIs of three specimens characterized as having high (8.16 kN, first row), medium (4.07 kN, second row), and low (1.15 kN, third row) failure load, respectively, are shown on the leftmost column of the first three rows. Subsequent columns in these rows show the scaling index method (SIM) transformations achieved with different radii, R = {1, 2, 3, 4, 5} and scaling factor, SF = 1. Scaling indices (α) are color-coded according to the colorbar shown on the right. The (α) histograms for these ROIs for different SIM radii are shown in the fourth row; representations of these histograms using 19 quantiles are shown in the fifth row. Note that only the central slice of each VOI is shown here; however, the histograms and quantile curves represent the distribution of α-values within the entire VOI. As seen here, the histograms and quantile curves can be used to distinguish between femur specimens of different biomechanical strengths.
Fig. 4
Fig. 4
Overview of the experimental setup and methods used. The true failure load (FLtrue) was recorded from biomechanical tests after MDCT imaging. Images were then postprocessed to facilitate conversion of intensity values from Hounsfield units to BMD. Different trabecular bone features (statistical moments of MDCT BMD distribution, geometrical features from SIM, and morphometric parameters) were computed from VOIs annotated on the postprocessed MDCT images. Two function approximation methods, i.e., multiregression and support vector regression, were then used to predict FLtrue. Prediction performance was quantified using root mean square error (RMSE) and R2.
Fig. 5
Fig. 5
Prediction performance (RMSE) achieved with SIM-derived geometric features computed for different radii (R) and scaling factor (SF) values using both multiregression and support vector regression. Each RMSE distribution is represented by the central mark that corresponds to the median, and edges that correspond to the 25th and 75th percentile. As seen here, the best prediction performance is achieved for SF = 1 and R = 5, as marked with bold triangles.
Fig. 6
Fig. 6
Comparison of prediction performance (RMSE) for trabecular bone characterizing features extracted from the femoral head (top), neck (middle), and trochanter (bottom) on MDCT, i.e., mean BMD, SIM features, and morphometric parameters (Morph.), when processed with both multiregression and support vector regression. For each RMSE distribution, the central mark corresponds to the median and the edges are the 25th and 75th percentile. The best prediction performance was noted achieved with both regression models in the head (marked with *) when compared to the neck and trochanter (best features marked with +).

Similar articles

See all similar articles

Cited by 7 articles

See all "Cited by" articles

LinkOut - more resources

Feedback