IN VIVO PREDICTION OF BONE STRENGTH USING SPECTRAL ANALYSIS

Abstract

Introduction: Substantial evidence is now available that complex multi-variate models such as Artificial Neural Networks and Finite Element Analysis can predict bone strength better than DXA. In order to build such models effectively, it is essential to determine which basic individual parameters will be used. The current study attempts to improve a number of existing parameters that reflect bone structure, originating from spectral analysis of cancellous bone in radiographic images, to assess whether their correlation with mechanical strength of bone can be improved.

M& M: Sixty standard AP x-rays of cadaveric human radii, for which mechanical data was available, have been examined. The bones had been mechanically tested to destruction and the details of the test have been published previously. The x-rays were digitized at 160 mm/ pixel, using a dedicated scanner. ROI of 64x64 pixels corresponding to 1x1 cm in the original x-rays were used for the analysis. Low pass and High pass filters were moved stepwise to determine the most effective bandwidth for the identification and measurements of the magnitude peaks in the fast Fourier transform (FFT).

Results: The spectral trabecular index had a correlation with load at fracture (LF) of −0.002 and with the work at fracture (WF) of −0.07. The filtered parameter, termed spectral index of bone structure (SIBS), increased these correlations to 0.41 and 0.46 respectively. The Longitudinal trabecular index had a correlation of 0.09 with LF and 0.04 with WF. The corresponding filtered parameter, termed longitudinal trabecular index of bone structure (LIBS), increased these correlation coefficients to 0.39 with LF and 0.41 with WF. Finally the Transverse trabecular index had a correlation of −0.19 with LF and −0.04 with WF. The Transverse index of bone structure (TIBS) altered the correlations to 0.17 with LF and 0.36 with WF. For this sample size, the 5% significance threshold for correlations is 0.25 and for the 1% level is 0.325.

Discussion: This refinement of the individual spectral parameters is an essential step towards the improvement of multivariate models, leading to a potentially improved assessment of fracture risk. The general assessment of trabeculae and particularly the longitudinal ones was substantially improved by the new method of measurement. These parameters can now be incorporated into more complex models that take into account other characteristics such as age of the patients, cortical thickness and size of the bones and which are knowingly related to bone fragility.