Multivariate calibration for quantitative analysis of EDXRD spectra from a bone phantom.

Phantoms have been constructed to simulate trabecular bone mineral loss and cortical bone thinning which consist of a mixture of hydroxylapatite powder and animal fat in various quantities, surrounded by a varying thickness of dural sleeve. Energy dispersive X-ray diffraction (EDXRD) spectra have been recorded, and multivariate calibration has been performed on the spectra from the bone phantoms. The multivariate technique of partial least squares (PLS) was used to predict the hydroxylapatite content of the phantoms and the dural thickness for measurement times of 250, 50 and 5 s. The calibration phantoms consisted of 10 hydroxylapatite densities ranging from 0.5852 g cm-3 to 0.3703 g cm-3 representing a loss of hydroxylapatite of approx. 40% in 4% intervals. Each phantom had four dural sleeves of thickness 0.5, 1.0, 1.5 and 2.0 mm. Nine test phantoms were constructed with a range of densities that were inside the calibration range. For a measurement time of 250 s the average accuracy of prediction for hydroxylapatite content was approx. +/- 3% while for a measurement time of 5 s this fell to approx. +/- 8%. The dural thickness was predicted to within approx. +/- 0.25 mm for a measurement time of 250 s. The results show that multivariate calibration is a useful technique for obtaining quantitative data of a desired variable from EDXRD measurements which may otherwise be masked or corrupted by other variables.